feat: add distance scores to dense search results - #101

Summary

Search results now return a distance score alongside each point ID (lower = closer).

Changes

• VectorIndex::search returns (PointId, distance) pairs, updated Flat, KD-tree and HNSW
• HNSW returns true Euclidean distance (sqrt applied), matching Flat and KD-tree
• HNSW now rejects queries whose similarity differs from the index's
• Cosine distance returns 1.0 for zero vectors instead of NaN
• Python client: search / batch_search return (point_id, score) tuples