[codex] Improve extractor hot path performance

[Widthdom]

Summary

• Fix the C# reference-extraction slowdown by avoiding repeated value-receiver body rescans and expensive duplicate checks.
• Reduce non-C# extractor hot-path overhead by replacing candidate-by-candidate scans over growing symbol/reference lists with keyed HashSet/cache lookups.
• Add large-fixture runaway guards for C#, Rust, Shell, Go, Java, Kotlin, C++, Dockerfile, JavaScript, TypeScript, and Swift paths.
• Document the performance issue and prevention guidance in changelog/developer documentation.

Root Cause

Large generated files could trigger super-linear extraction work when supplemental extractors repeatedly scanned already accumulated references, symbols, or scan targets while processing each new candidate. The observed severe regression was in C# reference extraction; the follow-up audit found similar theoretical hot paths in several non-C# symbol/reference extractors.

Validation

• dotnet build src/CodeIndex/CodeIndex.csproj --no-restore -p:UseSharedCompilation=false -m:1 -v:minimal
• dotnet build tests/CodeIndex.Tests/CodeIndex.Tests.csproj --no-restore -p:UseSharedCompilation=false -m:1 -v:minimal
• Reflection harness for targeted extractor performance/correctness tests: 19 PASS
• dotnet run --project tools/CodeIndex.Changelog -- check
• git diff --check
• cdidx index ... --json
• cdidx searches confirmed no remaining symbols.Any, symbols.First, targets.Any, references.Any, or references.First hot-path matches in extractor code.

Note: dotnet test could not run in this environment because VSTest/MSBuild socket or pipe creation failed with Permission denied; targeted tests were run through the reflection harness instead.