Fix file count and symbol query normalization performance

[Widthdom]

Summary

• Scope file metadata count aggregation to the candidate file set before joining symbol/reference counts.
• Reuse normalized symbol query lists across recursive multi-query search/count paths.
• Add focused DbReader coverage for grouped count SQL and normalized query reuse.

Validation

• dotnet build CodeIndex.sln -p:UseSharedCompilation=false
• dotnet test tests/CodeIndex.Tests/CodeIndex.Tests.csproj --filter "FullyQualifiedName~DbReaderTests" -p:UseSharedCompilation=false
• dotnet run --project tools/CodeIndex.Changelog -- check
• adversarial review: No blocking/actionable issues found.

Note: an earlier full dotnet test CodeIndex.sln -p:UseSharedCompilation=false run hit a transient ConsoleCaptureTests.Capture_RestoresConsoleStreams_WhenActionThrows failure and then hung; the failing test passed when rerun directly.

Documentation and Changelog

• Added changelog.d/unreleased/1827.fixed.md
• Added changelog.d/unreleased/1954.fixed.md

Follow-up Candidates

• Existing issue Full index can stall in symbol extraction and exit with E013_INDEX_EXTRACTION_STALLED #2744 tracks the E013_INDEX_EXTRACTION_STALLED full-index stall observed while refreshing the local index.

Fixes #1827
Fixes #1954