fix(yaml): rowan parser handles flush-left block sequences + links loud-fail (REQ-091, v0.13.1)

[avrabe]

Summary

Two REQ-091 fixes that together close the F2 silent-failure where the default `rivet validate` (salsa + rowan-yaml) silently graded zero links for artifacts written in YAML's flush-left list style.

The CST bug — flush-left sequences silently dropped

`yaml_cst::parse_mapping_entry` required `child_indent > entry_indent` when descending into a same-line `value`. YAML allows the zero-indent block sequence form where the dashed list sits at the same column as the parent key:

```yaml
artifacts:

• id: A # <-- dash at column 0, same column as "artifacts:"
  type: req
  ```

For this shape the parser silently treated the value as empty and dropped the entire sequence from the tree. `extract_schema_driven` then returned 0 artifacts + 0 diagnostics, while `rivet list` (which routes through the legacy `parse_generic_yaml`) returned the artifact correctly — the link graph the validator should grade was invisible.

The fix: when the next-line content is a `Dash`, accept `child_indent >= entry_indent` (YAML 1.2's zero-indent block-sequence rule).

CST probe — both shapes now produce identical trees:
`Mapping → MappingEntry → Value → Sequence → SequenceItem → Mapping`.

The F2 silent-failure — `extract_links_via_serde` silently returned empty

The fallback for `links:` values the CST doesn't recognise as a block `Sequence` (most importantly flow-style `links: [{type:X, target:Y}]`) silently returned `Vec::new()` when `serde_yaml::from_str` rejected the value text. The outer cardinality validator then graded that as "links field present but empty."

Threads `&mut Vec` through `extract_links` and `extract_links_via_serde`. On serde parse error, emit an ERROR diagnostic naming the underlying error and the value's span.

Tests

• New: `rivet-core/tests/req_091_flush_left_yaml.rs` — drives `extract_schema_driven` with both flush-left and indented fixtures; asserts 1 artifact + 1 link parity (Acceptance Add ASPICE 4.0, ReqIF, OSLC, WASM, and CI pipelines #1 + docs: add phase 2.5 features and CLI tests #4).
• New: `yaml_hir::tests::extract_links_via_serde_emits_diagnostic_on_parse_error` — malformed `links:` produces a diagnostic naming the field (Acceptance feat: AADL integration — schema, adapter, WASM rendering, architecture dogfood #3).

Test plan

• `cargo test -p rivet-core` — all tests pass (1068 lib + 83 yaml-test-suite + 2 new integration tests)
• `rivet validate` smoke test on this repo's corpus — PASS
• CST probe before vs after — flush-left now produces the same tree as indented
• CI

What this does NOT do

Acceptance #5 — re-grading the parallel agent's sphinx-needs port corpus against the Python reference's ~2500 warnings — requires their `feat/import-results-sphinx-needs` branch. Once this lands they can rebase and re-grade.

Fixes: REQ-091
Refs: REQ-051, REQ-082

🤖 Generated with Claude Code

[github-actions]

📐 Rivet artifact delta

No artifact changes in this PR. Code-only changes (renderer, CLI wiring, tests) don't touch the artifact graph.

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Rivet Criterion Benchmarks'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: 798c5d7	Previous: 54d609a	Ratio
store_lookup/100	2178 ns/iter (± 10)	1688 ns/iter (± 3)	1.29
store_lookup/1000	26885 ns/iter (± 136)	19128 ns/iter (± 836)	1.41
traceability_matrix/1000	60540 ns/iter (± 238)	41367 ns/iter (± 727)	1.46
query/100	797 ns/iter (± 4)	650 ns/iter (± 19)	1.23
query/1000	7621 ns/iter (± 35)	5409 ns/iter (± 32)	1.41

This comment was automatically generated by workflow using github-action-benchmark.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!