docs: add scoring & ranking deep dive to recall docs

[nicoloboschi]

Summary

• Adds a new Scoring & Ranking Deep Dive section to the recall retrieval docs
• Documents the full pipeline: RRF fusion formula, cross-encoder reranking, combined scoring boosts (recency, temporal proximity, proof count), recall budget mapping, and graph scoring detail
• Includes design rationale ("why") for each algorithm choice, drawn from code comments
• Removes the now-redundant "Graph Retrieval Algorithms" section (only one algorithm exists)
• Updates both docs/ and versioned_docs/version-0.5/

Test plan

• Docs build passes in CI
• Verify rendered page at Developer → Recall → Scoring & Ranking Deep Dive