Optimize InternalNames lookup

[mcculls]

What Does This Do

Replace bounded rehashing and repeated table growth in InternalNames with a fixed table sized to at most 50% load. Resolve collisions with linear probing and cache hashes in a parallel array so lookups skip strings whose hashes differ. Names are normalized and deduplicated during construction, and lookups continue to support allocation-free queries using String or any CharSequence with a String-compatible hash code.

Motivation

Distinct names with identical String hashes follow the same probe sequence. Repeatedly growing the table cannot resolve those collisions and can exhaust the heap. Linear probing handles these collisions without rebuilding or growing the table.

Additional Notes

• Add regression coverage for identical hashes, probe runs and wraparound, normalized duplicates, missing names, allocation-free CharSequence queries, matcher factory integration, and set behavior.
• Move the shared class-name corpus into testing, expose it through SampleClasses.loadClassNames(), and update the utils benchmarks to use it.
• Add an InternalNames benchmark that checks all 33,957 corpus names against 25 groups sampled from dd-trace-java with both String and TypeString queries. It reports time per corpus class across all groups, normalizes names and precomputes query hashes outside measurement, validates membership against HashSet during setup, and returns the total match count to keep all lookups observable.

Contributor Checklist

• class-match tests passed on JDK 8, 11, 17, 21, and 25.
• class-match formatting checks passed.