benchmarks: add retrieval harness and docs

[shadow6427]

Closes #24

This PR implements the retrieval benchmark suite described in Phase 3 of the roadmap.

What's included:

• scripts/bench/ Harness: An automated Python benchmarking suite (main.py) that evaluates Recall@K (1, 5, 10) and p50/p95 latency.
• Dataset Generation: Uses the HuggingFace datasets library to pull a subset of BEIR (scifact by default) and compute embeddings using sentence-transformers.
• Database Clients: Pluggable adapters (clients.py) for:
  • Engram (EngramBenchClient)
  • Pinecone (PineconeBenchClient)
  • Weaviate (WeaviateBenchClient)
  • pgvector (PgVectorBenchClient)
• Docker Compose: Includes scripts/bench/docker-compose.bench.yml to effortlessly spin up local instances of Weaviate and PostgreSQL(pgvector).
• Documentation: Automatically generates a docs/benchmarks.md markdown report when run.

Running the harness:

cd scripts/bench
docker compose -f docker-compose.bench.yml up -d
python main.py

(To generate a mocked version of the docs, run python main.py --mock)

[vercel]

@shadow6427 is attempting to deploy a commit to the praise's projects Team on Vercel.

A member of the Team first needs to authorize it.