Build comparative performance benchmarks and GitHub regression history

## Outcome

Measure standalone native Tantivy and complete Harper-derived native indexing, with a no-index Harper control and release history.

## Acceptance criteria

- Use deterministic corpora, analyzers, mutations, publication cadence and queries with exact result parity.
- Measure standalone wrapper, Harper without fulltext, and Harper with native derived indexing. These are usage paths/control, not separate storage backends.
- Distinguish native engine measurements from Table.search/REST product latency; integrated differences include source writes, projection, replay, auth and record retrieval.
- Report ingest/update/delete throughput, per-query-class p50/p95/p99, source/replica latency, indexing lag/backlog drain, errors, packing, native queue/execution, commit/reload, CPU/RSS/mappings and disk/I/O.
- Include heavy-tail text, warm/cold caches, realistic filters, fixed-arrival and saturation load, multiple indexes, unrelated-only source writes, replicas and merge/rebuild pressure.
- Exercise restart/replay and initial rebuild; distinguish process-kill recovery from power-loss durability qualification.
- PR CI gates correctness/output schema; scheduled/release profiles use controlled hardware and reviewed compatible-cohort thresholds.
- Store versioned JSON and immutable environment/build/workload fingerprints in GitHub for release comparisons; redact machine paths/source text/raw errors.
- Keep old hosted/KV/RocksDB results labeled historical. No RocksDB storage comparison or third-source benchmark is a release gate.

## Design

[Native Tantivy storage and Harper derived indexes](https://github.com/HarperFast/fulltext/blob/codex/native-storage-design/docs/native-storage-integration.md). Native Tantivy files are the only delivery target for standalone use and Harper. Proposed APIs and integration gates are not claims of completed implementation.

Comment generated by kAIle (GPT-6)