feat: add benchmark artifact tooling

[raoabinav]

Evaluator Summary

• Abi added reproducible benchmark artifact tooling for recall, latency, storage, provenance, query-log summaries, and backend comparisons, moving LEANN evaluation from ad hoc scripts toward auditable reports.
• The design records input hashes, timing/storage stats, provenance, and machine-readable JSON plus Markdown summaries so benchmark runs can be compared across datasets and backends.
• The implementation includes shared metrics/provenance/storage helpers and hardens BM25/DiskANN baseline scripts so future benchmark work has a consistent artifact contract.
• Quality bar: unit tests cover benchmark generation, backend comparison manifests, query-log summaries, metrics, provenance, and baseline scripts; lint/format and diff checks were run.

[raoabinav]

Replaced by #357 from the neutral feature branch.