v0.5.1 — packaging + docs corrections

Packaging + docs corrections on top of 0.5.0. No library code changes — retrieval behavior is identical to 0.5.0. If you're on 0.5.0 this is a safe, recommended bump.

Changed

• Packaging metadata intended for the 0.5.0 cut: Development Status 3 - Alpha → 4 - Beta; PyPI Homepage + Documentation → attune-rag.dev.
• README now leads with the measured numbers — bundled corpus 100% precision@1 / 100% recall@3 / 0.97 faithfulness (CI-gated); on an unseen corpus the optional transformer tier takes hard-paraphrase precision@1 25% → 90% and recall@3 → 100%; abstention cuts out-of-corpus false answers 92% → 8%. Adds a lightweight-vs-transformer guide, corrected install instructions ([embeddings] + [transformers] tiers), and an accurate runtime-dependency count (5 pure-Python; footprint scales by setup).
• attune-rag.dev landing page + help corpus refreshed to the current numbers and the 0.5.0 retrieval code.

Install

pip install attune-rag             # core — keyword retrieval, 5 pure-Python deps, no LLM SDK
pip install 'attune-rag[embeddings]'   # + torch-free static hybrid retrieval
pip install 'attune-rag[transformers]' # + transformer tier (pulls torch, ~GB)

Docs & guides: https://attune-rag.dev