Wire IndicatorValidator + RecordStructureValidator at strict_marc (bd-0x73.1)

[dchud]

Summary

Resolves bd-0x73.1 — the four unwired validator structs from the v0.8
audit. Decisions per validator:

• IndicatorValidator → wired at validation_level="strict_marc" (per-tag MARC 21 indicator rules; IsbnValidator/EncodingValidator left opt-in).
• RecordStructureValidator → wired at validation_level="strict_marc" (MARC 21 leader-byte semantics).
• IsbnValidator → opt-in callable API; documented in new docs/reference/validators.md.
• EncodingValidator → opt-in callable API; documented alongside IsbnValidator. Heuristic detection deliberately stays out of strict_marc since strict_marc must be deterministic.

The principled split is format-semantic vs. content/heuristic, not a BC-driven carve-out (pre-1.0).

Design call worth flagging

Per-tag indicator violations and leader-byte semantic violations reuse existing error codes E201 and E002 with tag- or position-specific expected: strings, rather than minting new variants. Granularity belongs at validation_level, not at the error-code dimension — same conceptual concern, same code, different sub-rule. This keeps the public error surface from fragmenting as more validators get wired.

Both new failure shapes are recoverable in lenient/permissive, consistent with bd-0x73.10's single rule.

RecordStructureValidator::validate_leader now fires InvalidLeader (E002) instead of the misleading InvalidField it built before (drive-by fix).

Test plan

• cargo test --test validation_level_matrix — two new fixtures (e201_per_tag_indicator_245.bin, e002_invalid_record_status.bin) exercising the 2×3 {validation_level} × {recovery_mode} matrix.
• tests/error_coverage.toml — both cases registered with validation_level = "strict_marc"; harness reports 20/20 wired (was 18/18).
• .cargo/check.sh (full): 538 lib/integration + 32 doc tests pass; clippy clean (core + python); ruff clean; maturin builds; mkdocs builds.
• Python suite: 731 passed / 8 skipped.

Docs

• New docs/reference/validators.md page covering all four validators with auto-run vs opt-in classification, code examples, and the format-semantic / content-heuristic principle.
• docs/reference/error-codes.md — E002 and E201 entries describe both failure shapes.
• docs/reference/error-handling.md — validation-level table extended with the per-tag and leader-semantic rows.
• CHANGELOG.md [Unreleased] updated.

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks
🆕 2 new benchmarks
⏩ 60 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	Simulation	parse_10k_clean_strict	N/A	90.5 ms	N/A
🆕	Simulation	parse_10k_bad_indicators_lenient	N/A	90.3 ms	N/A

---

_{Comparing feat/bd-0x73.1-validators (e60895f) with main (deacc24)}

Footnotes

1. 60 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩