Build comparative performance benchmarks and GitHub regression history

[kylebernhardy]

Outcome

Measure the cost and scalability of the wrapper, especially RocksDbDirectory relative to the native Tantivy reference.

Acceptance criteria

• Paired workloads compare @harperfast/fulltext/native and @harperfast/fulltext/rocks using identical corpora and queries.
• Measurements include ingest throughput, commit/reload latency, search p50/p95/p99, memory, cache behavior, write amplification, storage size, recovery, and rebuild.
• Query classes are reported separately for term, Boolean, phrase, prefix, fuzzy, autocomplete, highlighting, and filtered search.
• Multi-index and mixed foreground RocksDB workloads are included.
• CI runs stable smoke gates on pull requests and scheduled/release benchmark suites on controlled hardware.
• Machine-readable results and summaries are retained in GitHub and compared release over release with environment metadata and regression thresholds.

Comment generated by kAIle (Codex GPT-5)