feat: normalize search scores to 0–1 similarity range

[daniel-harperdb]

Summary

Adds similarity score normalization to MemorySearch and SynapseSearch results.

• Converts HNSW cosine $distance (range 0–2) to similarity (range 0–1) via Math.max(0, 1 - $distance / 2)
• Adds similarity field alongside existing $distance on each result
• Required for LangChain compatibility (expects 0–1 similarity scores)

Stacked on #7 (memory-count).

Test plan

• Unit tests included (test/score-normalization.test.js)
• Manual test: verify similarity=1.0 for exact match, ~0 for distant vectors
• Verify $distance field is still present for backwards compatibility