Cardano Updates

This are deliberately vague, but I'm trying to capture some high-level
ideas of being as concrete as possible and have little conditional wiring.

Reparameterise Cardano.Api.Experimental.SignedTx and its
SerialiseAsRawBytes instance on ShelleyLedgerEra era instead of
LedgerEra era. This lets SignedTx and its raw-bytes deserialisation
type for every Shelley-based era (Shelley through Dijkstra) without
widening the LedgerEra family or its Conway-onwards constraint
surface (EraCommonConstraints), which would have rippled into the
experimental tx builders.

Build paths (signTx, makeKeyWitness, convertTxBodyToUnsignedTx) take
Era era, whose constructors only cover Conway/Dijkstra, so they
remain Conway+ gated; for those eras LedgerEra era ~ ShelleyLedgerEra
era is provided by obtainCommonConstraints, so the SignedTx
constructor application elaborates with no further changes.

The motivation is unblocking cardano-cli's debug transaction view,
which handles InAnyShelleyBasedEra and needs to deserialise into
Exp.SignedTx era for pre-Conway tx files.

Signed-off-by: Chris Gianelloni <[email protected]>

Iteration-3 fixes verified by Codex:
- synthesize_fail always writes status: fail when result.json is missing
- FAILURES counter correctly reflects synthesis-flipped verdicts
- Authored result.json is parse-validated before move; invalid JSON
  falls through to synthesize_fail
- No new blockers introduced

One cosmetic polish applied post-convergence: captured the adapter's
numeric exit code (cell_exit) and used it in the synthesized error
message (`adapter exited (exit code N) without writing result.json...`)
instead of the verdict word. Not structural; not a blocker.

Final state: 4 iterations to convergence; 6 substantive issues
addressed across iterations 1–3. The CI integration of the primitives
matrix into ecosystem-test.yml is deferred (deliberate, documented);
scripts/run-conformance.sh runs locally today and writes results in
the aggregator's expected shape so the CI wiring is mechanical when
adapter impls are validated against working SDK environments.

.codex-reviews/p2w1.md: iteration 4 verdict + final state recorded.

Milestone status: ready for human review.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Iteration 2's synthesis path closed the original disappearing-cell bug
but introduced two new gaps Codex caught:

(P2.1) Zero-exit-no-result false PASS: if an adapter exited 0 but failed
to write result.json, the synthesized result took status from cell_status
(pass) — but missing result.json IS a contract violation regardless of
exit code. Fixed: replaced the synthesis branch with a synthesize_fail
helper that ALWAYS writes status: "fail" when result.json is missing,
and increments the FAILURES counter when synthesis flips pass → fail
so the runner's overall exit reflects reality.

(P2.2) Corrupt JSON disappearance: an adapter that wrote a partial or
unparseable result.json was blindly moved to .ci-results/, then silently
skipped by the aggregator (which `continue`s on parse failure). Fixed:
after the adapter writes result.json, the runner validates it parses as
JSON before moving; if it doesn't parse, falls through to synthesize_fail
with a clear error_summary pointing at the log file.

Both gaps were strict-mode side effects of iteration 2's strict-mode
design and don't change the milestone's deliberate deviations
(CI-wiring deferral remains documented).

.codex-reviews/p2w1.md: iteration 3 actions logged.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

NEW P1: scripts/run-conformance.sh only moved an adapter-authored
result.json. When Deno/JBang/import setup fails before the adapter code
runs, no result.json gets written and the cell silently disappeared from
matrix.json. This conflicts with the runner-contract guarantee that every
attempted cell produces a result entry — silent skips are forbidden.

Fixed: when an adapter exits without writing result.json, the runner now
synthesizes one with tier: primitive, scenario id, primitive id, era,
framework, status: fail, captured duration_ms, and error_summary
pointing at the log file. The cell now appears in matrix.json regardless
of where in its lifecycle the adapter failed.

NEW P3: conformance/README.md:59 said "lands in P3 alongside use-case
scenarios" — stale after iteration 1 added the doc. Rewrote the link
description to reflect that Tier 1 is live in P2W1 and Tier 2 is
stubbed pending P3.

.codex-reviews/p2w1.md: iteration 2 actions logged.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>