Decompose large indexing and query orchestration

[Widthdom]

Summary

• Decompose large symbol extraction, reference extraction, database query, MCP, and CLI implementations into cohesive partials and helpers.
• Consolidate mutable ownership in update and full-scan orchestration, removing duplicate context/result state and callback handoffs.
• Preserve behavior with characterization and parity coverage while fixing four correctness defects found during the decomposition.
• Reduce hot-path allocations in pattern extraction and JVM triple-string masking.

Why

Several core paths had grown into very large classes and methods with duplicated mutable state, callback-heavy context objects, and difficult lifetime boundaries. This series makes those boundaries explicit while preserving transaction, cancellation, schema-fallback, and output behavior.

The branch contains 45 focused commits so each extraction or correction remains independently reviewable.

Correctness fixes

• Use one grammar-aware C# base-list parser so generic constraints and nested type syntax are resolved consistently.
• Replay accepted but undispatched parallel batch inputs after an output-limit stop instead of losing them.
• Apply C# partial-family unused-symbol filtering before paging without per-candidate queries, while preserving legacy read-only schema fallbacks.
• Keep full-scan FTS mutation state monotonic and defer shared worker-resource cleanup until all peer workers stop.

No open issue is auto-closed by this PR.

Validation

• Release solution build: 0 warnings, 0 errors.
• Full Release test suite, net8.0: 11,395 passed, 7 skipped, 0 failed.
• Full Release test suite, net9.0: 10,924 passed, 417 skipped, 0 failed.
• Full-scan focused suite: net8.0 181 passed / 1 skipped; net9.0 179 passed / 3 skipped.
• dotnet format CodeIndex.sln whitespace --no-restore --verify-no-changes
• dotnet run --project tools/CodeIndex.Changelog -- check — 4 fragments validated.
• git diff --check origin/main..HEAD
• Dogfood index rebuilt at HEAD with complete index/reference graph and no warnings or errors.
• Two independent adversarial review lanes reported no blocking or actionable findings.

Documentation and changelog

Behavioral contracts were documented in the English and Japanese sections of TESTING_GUIDE.md.

Changelog fragments:

• changelog.d/unreleased/+batch-output-limit-replay.fixed.md
• changelog.d/unreleased/+csharp-base-list-parser.fixed.md
• changelog.d/unreleased/+full-scan-extraction-state.fixed.md
• changelog.d/unreleased/+unused-legacy-schema.fixed.md

Follow-up candidates

No correctness follow-up is required. Further decomposition of transaction-heavy full-scan/readiness tails and baseline per-file persistence carriers is intentionally deferred to separately characterized changes.