TxPlan (YAML) transaction building + Plutus + on-chain validation; toolchain & CI modernization

[matiwinnetou]

Summary

Brings develop to main. Two bodies of work have accumulated since the last release: (A) toolchain/CI/docs modernization that got CI green and added developer docs, and (B) a large refactor that replaces the bespoke JSON transaction format with CCL's native TxPlan (YAML), adds Plutus support, and validates the whole intent surface against a real Cardano node.

Net: ≈−11,300 lines across ~113 files (59 commits) — the bespoke tx-spec, its mappers, the provider path, and the four per-language fluent builders are gone.

---

⚠️ Breaking changes (for existing consumers)

This release changes the transaction-building contract end to end. Any code on the old JSON-operations API must migrate.

	Before (main)	After (this PR)
Native entry point	ccl_quicktx_build(thread, specJson) — one bespoke JSON spec	ccl_quicktx_build(thread, yaml, utxos_json, protocol_params_json, exec_units_json)
Wrapper API	fluent TxBuilder / ScriptTxBuilder (.payToAddress(...).build()) + ProviderConfig	build(yaml, utxos, protocolParams, execUnits?) — pass a TxPlan YAML string + caller-supplied chain data
Chain data	fetched via provider (HTTP) or inlined in the spec	always caller-supplied (utxos, protocolParams); no provider/HTTP path
Result payload	JSON {tx_cbor, tx_hash, fee}	YAML {tx_cbor, tx_hash, fee}
Signing stake/gov txs	n/a (payment key only)	ccl_account_sign_tx is payment-only and unchanged; stake/DRep/committee certs now require the new sign_tx_with_keys(..., ["payment","stake"]) or they're rejected with MissingVKeyWitnessesUTXOW

# Before
b = lib.quicktx.builder()
b.pay_to_address(to, amount)
result = b.build()                      # JSON out

# After
result = lib.quicktx.build(             # YAML out
    txplan_yaml, utxos, protocol_params)

New wrapper runtime dependencies (a YAML parser per wrapper): Python pyyaml>=6.0 · Go gopkg.in/yaml.v3 · Rust serde_yaml 0.9 · JS yaml ^2.3.0.

---

Part A — Toolchain, CI & docs

1. Standardize on Oracle GraalVM 25.0.3 — align local build + docs with CI's distribution: 'graalvm'; pin CI/release to the exact patch (was floating '25') for reproducibility.
2. Fix CI toolchains — add actions/setup-go + dtolnay/rust-toolchain (the matrix lacked Go/Rust, so those tasks failed with command not found); this got macOS green.
3. Fix the Go wrapper's thread-affinity bug — the Go suite crashed on Linux x86_64 with a GraalVM "yellow zone" StackOverflowError (a goroutine migrating off the OS thread that created the isolate). All FFI calls now run on one dedicated OS thread (runtime.LockOSThread + an executor goroutine) for the Bridge's lifetime.
4. Build libccl.dll on Windows — new CI Windows job + a windows-x86_64 release artifact (DLL + import library + headers).
5. DevKit integration-test runner — integration-tests.yml starts Yaci DevKit (admin API :10000) and runs the wrappers' integration suites (real build → sign → submit). (First introduced here; now green — see Part B.)
6. Per-wrapper READMEs + runnable offline examples, and a categorized, prioritized TODO.md backlog (incl. an upstream-CCL scan and a Non-Goals section). CI also now runs on PRs into develop.

---

Part B — TxPlan (YAML) transaction building

7. Upgrade CCL 0.7.2 → 0.8.0-pre4 — backward-compatible for the bridge; only the deprecated ScriptTx path changes.
8. Core: TxPlan in, YAML out — deleted the bespoke spec + mappers + provider path (~2,400 LOC). ccl_quicktx_build(thread, yaml, utxos_json, protocol_params_json, exec_units_json) builds fully offline from a TxPlan YAML document with caller-supplied chain data; the result {tx_cbor, tx_hash, fee} is YAML.
9. Plutus script transactions (offline) — execution units are a fourth caller-supplied input; the bridge wires StaticTransactionEvaluator to stamp them on (it never runs the script). Includes the native-image reflection registration the Plutus/cert/serializer classes need.
10. Stake / DRep-key signing — ccl_account_sign_tx_multi(…, keys) signs with any subset of payment / stake / drep / committee_cold / committee_hot, fixing MissingVKeyWitnessesUTXOW for stake/vote/DRep certs (the original payment-only ccl_account_sign_tx is unchanged).
11. Wrappers — thin YAML pass-through — deleted the fluent builders (~10k LOC across Python/Go/Rust/JS); each exposes build(yaml, utxos, protocolParams, execUnits?) + sign_tx_with_keys.
12. Intent coverage, every wrapper, end-to-end — a shared, version-controlled fixture set (test-fixtures/quicktx-intents/, generated by the JVM from CCL's exact intent shapes) driven through the native library in Go, Python, Rust, JS: payments, metadata, native mint, donation, staking, DRep, voting, governance proposals, pools, native scripts, reference/explicit inputs, compose (multi-sender), and both Plutus paths.
13. On-chain validation (DevKit) — the Go integration suite now does build → sign (right key roles) → submit → assert on-chain for the intent set, with the required setup sequenced (register a stake address before a proposal; register a pool before delegating; vote-delegate before withdrawing). Depth: mints assert the asset landed; the Plutus spend asserts the script UTXO was consumed; a negative test proves a payment-only-signed DRep registration is rejected.

---

Verification

• :core:test + :core:nativeCompile + :native-test:test green on Oracle GraalVM 25.0.3.
• All four wrapper suites green (per-intent E2E + Plutus).
• CI green on Linux x86_64, macOS ARM64, Windows.
• Integration Tests (DevKit) green — the full intent set is submitted to and accepted by a live devnet.

Known limitations / follow-ups (tracked in TODO.md, not blockers)

• Pluggable exec-unit evaluators (§2b): client-side Blockfrost/Ogmios/Aiken/Scalus helpers for obtaining the units to pass in.
• Cross-wrapper submit: only Go submits to DevKit (the CBOR is produced by the shared native lib, so Go-only proves node-acceptability; the other wrappers have build E2E parity).
• Windows wrapper test coverage beyond the DLL build.