Rework README around benchmark evidence

[ponythewhite]

Summary

• Rewrites README.md around auditable benchmark evidence, explicit API contract, auto selector behavior, hardware/runtime settings, dataset matrix, FAISS settings, timeout/pruning policy, and reproduction commands.
• Adds raw real/ANN/synthetic benchmark JSON snapshots plus generated README summary CSVs.
• Tightens two static auto selector guardrails so completed measured auto rows stay within the documented 2.5% quality window.
• Bumps package version to 1.0.5 and removes standalone Markdown docs so only README.md and CONTRIBUTING.md remain.

Verification

• python scripts/summarize_benchmark_evidence.py
• python -m pytest -q
• python -m py_compile scripts/summarize_benchmark_evidence.py
• cargo check --no-default-features --features python
• cargo test --no-default-features --features python
• git diff --cached --check before commit