Cardano Updates

Move the run/dmtrcli logic out of e2e.yaml into .github/e2e/scripts/
(run-test.sh, setup-dmtr-socket.sh, stop-dmtr-tunnel.sh), invoked from the
workflow steps. Scripts read the same env vars the steps already set; committed
with the executable bit so they run after checkout.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Replaces the EKS-backed e2e workflow with one that runs entirely on
github-hosted runners. Each leg runs the release image via `docker run` +
`timeout` and gates on the process exit code; the new WorkStats filter gives
each test a finalization policy so oura self-terminates after N blocks
(exit 0), the Assert sink panics on a bad block (non-zero), and `timeout`
guards a hang (124).

- delete the `prepare` job (EKS kubeconfig / namespace / IRSA service account)
- extract each config from its k8s ConfigMap into .github/e2e/configs/*.toml,
  rewritten for the oura v2 schema (peers/socket_path, top-level
  chain/intersect, LegacyV1 filter for the Assert sink, SplitBlock+ParseCbor
  for cloud sinks, RollbackBuffer for min_depth, WorkStats for finalization)
- AWS creds via aws-actions/configure-aws-credentials (assuming the
  OuraE2ETest role) injected into the container instead of EKS IRSA
- n2c uses a Demeter dmtrcli socket port-forward (placeholder; DMTR_API_KEY
  secret) and is continue-on-error until that path is finalized

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The U5C source connects over TLS via tonic/rustls. In builds that pull more
than one rustls crypto provider (e.g. the `aws` feature brings in aws-lc-rs
alongside ring), rustls has no process-default provider and panics on the
first TLS handshake (`no process-level CryptoProvider available`). Install the
ring provider once at startup, gated on the `u5c` feature.

This only surfaced now because nothing exercised U5C in a multi-provider build;
the manual u5c tests built `--features u5c` alone (single provider).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The three relay.cnode-m1.demeter.run endpoints map to different networks:
:3000 = Mainnet, :3001 = PreProd, :3002 = Preview. Mainnet legs use :3000,
preview n2n legs use :3002 (replacing the public IOG/world relays).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Move the manual u5c checks into the e2e suite as three `kind: u5c` legs
against the Demeter utxorpc endpoints, using the standard exit-code gate via
a WorkStats finalization policy (3 blocks -> exit 0); a #921-style regression
(reset loop, no apply events) never finalizes -> timeout -> fail.

API keys are parameterized as ${DMTR_UTXORPC_KEY_*} and resolved by envsubst
from GitHub secrets.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Bumps gasket 0.7 -> 0.10, which brings the non-panicking teardown and the new
`TetherState::Finished(StagePhase)` (PR construkts/gasket-rs#35). Adapt to the
0.10 API:

- `Daemon(tethers)` -> `Daemon::new(tethers)`.
- `block`/`teardown` now consume `Daemon`. `dump`/`watch` just call `block`
  (which tears down on stop). `run_daemon` shares the daemon via `Arc` with the
  prometheus + tui tasks, so it drives the stop loop over `&self` and reclaims
  ownership (after the aborted tasks drop their clones) for a graceful teardown.
- `console` handles the new `Finished` tether state.

Crucially, in 0.10 a crashing stage no longer aborts the process (gasket
isolates it per-thread), so `run_daemon` now derives the exit code from the
stages' terminal state: a stage that reached `Ended` (a finalization filter
completed) or a user termination signal -> exit 0; otherwise a stage crashed or
stalled -> non-zero. This is what makes the e2e exit-code gate meaningful: a
WorkStats finalize exits 0, an Assert panic exits non-zero.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Introduces a pass-through `WorkStats` filter that tracks work stats and
enforces a finalization policy (until_hash / max_block_slot /
max_block_quantity). When the policy is reached the filter ends its stage,
which gracefully stops the whole daemon (gasket halts once any stage reaches
`Ended`).

Because it lives in the filter chain it is decoupled from any particular
source — restoring the v1 source `finalize` semantics for every source at
once, and re-enabling the previously commented-out `max_block_quantity`
(the block counter now lives naturally in the filter). The name leaves room
to grow into a home for richer progress/throughput stats; `block_count` and
`latest_slot` metrics are the seed.

Reuses framework::FinalizeConfig / should_finalize (fields made public,
quantity branch re-enabled) and adds unit tests for the policy.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Move the manual tmp_u5c_test checks into the streamlined e2e suite as three
`kind: u5c` legs against the Demeter utxorpc endpoints. Converts the old
time-bounded run.sh approach to the standard exit-code gate via a WorkStats
finalization policy (3 blocks → exit 0); a #921-style regression (reset loop,
no apply events) never finalizes → timeout → fail.

API keys are parameterized as ${DMTR_UTXORPC_KEY_*} (resolved by envsubst from
GitHub secrets) rather than hardcoded — the scratch files held real-looking
keys and this repo is public. The e2e image is built --all-features, so U5C is
available (and now TLS-safe per the preceding crypto-provider fix).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The U5C source connects over TLS via tonic/rustls. In builds that pull more
than one rustls crypto provider (e.g. the `aws` feature brings in aws-lc-rs
alongside ring), rustls has no process-default provider and panics on the
first TLS handshake (`no process-level CryptoProvider available`). Install the
ring provider once at startup, gated on the `u5c` feature.

This only surfaced now because nothing exercised U5C in a multi-provider build;
the manual u5c tests built `--features u5c` alone (single provider).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Move the run/dmtrcli logic out of e2e.yaml into .github/e2e/scripts/
(run-test.sh, setup-dmtr-socket.sh, stop-dmtr-tunnel.sh), invoked from the
workflow steps. Scripts read the same env vars the steps already set; committed
with the executable bit so they run after checkout.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The three relay.cnode-m1.demeter.run endpoints are different networks, not
redundant mainnet relays:
  :3000 = Mainnet, :3001 = PreProd, :3002 = Preview

Point the mainnet legs at :3000 (only) and the preview n2n legs at :3002,
replacing the public preview relay. (:3001/PreProd is unused — no preprod legs.)

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Replace the IOG public mainnet relay with the three Demeter relay endpoints
(relay.cnode-m1.demeter.run:3000-3002) across the 5 mainnet n2n legs
(assert-n2n-mainnet-{origin,tip}, aws-{s3,sqs,lambda}).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Introduces a pass-through `WorkStats` filter that tracks work stats and
enforces a finalization policy (until_hash / max_block_slot /
max_block_quantity). When the policy is reached the filter ends its stage,
which gracefully stops the whole daemon (gasket halts once any stage reaches
`Ended`).

Because it lives in the filter chain it is decoupled from any particular
source — restoring the v1 source `finalize` semantics for every source at
once, and re-enabling the previously commented-out `max_block_quantity`
(the block counter now lives naturally in the filter). The name leaves room
to grow into a home for richer progress/throughput stats; `block_count` and
`latest_slot` metrics are the seed.

Reuses framework::FinalizeConfig / should_finalize (fields made public,
quantity branch re-enabled) and adds unit tests for the policy.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Replaces the EKS-backed e2e workflow with one that runs entirely on
github-hosted runners. Each leg runs the release image via `docker run` +
`timeout` and gates on the process exit code; the new WorkStats filter gives
each test a finalization policy so oura self-terminates after N blocks
(exit 0), the Assert sink panics on a bad block (non-zero), and `timeout`
guards a hang (124).

- delete the `prepare` job (EKS kubeconfig / namespace / IRSA service account)
- extract each config from its k8s ConfigMap into .github/e2e/configs/*.toml,
  rewritten for the oura v2 schema (peers/socket_path, top-level
  chain/intersect, LegacyV1 filter for the Assert sink, SplitBlock+ParseCbor
  for cloud sinks, RollbackBuffer for min_depth, WorkStats for finalization)
- AWS creds via aws-actions/configure-aws-credentials (assuming the
  OuraE2ETest role) injected into the container instead of EKS IRSA
- n2c uses a Demeter dmtrcli socket port-forward (placeholder; DMTR_API_KEY
  secret) and is continue-on-error until that path is finalized

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The Windows `Check` CI job failed compiling `[email protected]`'s C feature
tests (`c11.c`) under the runner's new MSVC toolchain (VS 18 / 14.51), which
GitHub is rolling out as it redirects windows-latest to windows-2025-vs2026.

Bump the rustls crypto backend, which transitively pulls a newer aws-lc-sys
that supports current compilers:

- aws-lc-rs  1.13.3 -> 1.17.0
- aws-lc-sys 0.30.0 -> 0.41.0

Lockfile-only change (transitive deps; the aws-sdk crates were already
current). Verified `cargo build --features aws` compiles locally; Windows
build is validated by CI.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The Lint job historically died at the cargo fmt step, so clippy never ran
and these latent lints accumulated unnoticed. Now that fmt passes, clear
them so `cargo clippy --all-features --all-targets -- -D warnings` is green:

- framework: derive `Default` for `ChainConfig` instead of a manual impl.
- select filter: use an array literal instead of `vec!` in a test.
- redis sink: remove the unused `StreamStrategy` enum.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

- pallas 1.1 deprecated the `MintedDatumOption` alias; use `DatumOption`
  in legacy_v1 crawl/map filters.
- the Nullable->Option change made two pool-metadata matches into manual
  `Option::map`; rewrite with `.as_ref().map(...)`.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

The U5C source against current UTxORPC servers (e.g. Demeter) emitted one
`reset` event then looped on `x=grpc error` with no apply events. The cause
was a breaking change in the UTxORPC spec: oura was pinned to
`utxorpc-spec 0.15` (via `utxorpc` 0.10 and `pallas` 0.35) while servers now
speak `0.19`, in which `BlockRef.index` was renamed to `slot`.

Rather than decouple oura's parsed-data model from pallas (pallas owns the
`ParsedBlock`/`ParsedTx` types and the CBOR->spec mapper used by the whole
`select` filter and `parse_cbor`), bump pallas and utxorpc together so both
ride a single `utxorpc-spec 0.19`:

- `utxorpc` 0.10 -> 0.14, `pallas` 0.35 -> 1.1

Spec/API changes handled:

- `BlockRef.index` -> `slot` and `LiveTip::event()` now returns
  `Option<TipEvent>` (None = server closed the stream) in `sources/u5c.rs`.
  Also surface the real gRPC status via `inspect_err(... error!(?err))`
  before restarting, instead of the opaque "grpc error" that masked #921.
- `TxOutput.coin: u64` -> `Option<BigInt>` and `Asset.{output_coin,mint_coin}`
  -> `Asset.quantity: Option<asset::Quantity>`. Added `big_int_to_u64` /
  `asset_quantity_to_u64` helpers and routed the `select` predicate call
  sites and test fixtures through them.
- `LedgerContext` gained `get_slot_timestamp`; `NoOpContext` returns `None`.
- pallas 1.1: `Nullable<T>` -> `Option<T>` for `Relay`/`PoolMetadata` fields
  in `legacy_v1/map.rs`; `pallas::storage::hardano::immutable` ->
  `pallas::interop::hardano::storage::immutable` in `sources/mithril.rs`.

Verified: `cargo check --all-features` clean; default and `u5c` builds link;
`cargo test --features u5c` passes (18/18).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>