initial support: Add high-level retrieval pipeline (RRF + optional MMR)

[MagellaX]

PR description

• Introduces a non-generated retrieval module at src/zeroentropy/lib/retrieval.py:

  • search_documents and asearch_documents to query multiple collections and fuse results via Reciprocal Rank Fusion.
  • Optional Maximal Marginal Relevance diversification to reduce duplicates.
  • Returns UnifiedDocumentResult with fused score and per-collection provenance.

• Adds tests tests/test_retrieval.py validating multi-collection fusion (sync/async) with respx.

• Adds example examples/retrieval_basic.py demonstrating multi-collection retrieval.

Benefits: better retrieval quality out-of-the-box (recall + precision + diversity) without changing generated endpoints; lives in lib/ to avoid codegen conflicts.

[MagellaX]

Hey @npip99, any thoughts?