Performance optimization: select_context on large graphs

[jeremiepas]

Problem

Profiling needed for select_context with 100k+ node graphs. Current implementation uses Map lookups and Set operations which may degrade at scale.

Solution

Profile and optimize:

1. Benchmark select_context on 10k, 50k, 100k node graphs
2. Optimize hot paths (relevanceScore, community-aware selection)
3. Consider HNSW index for embedding search on >50k nodes
4. Cache index rebuilds only when graph changes

Acceptance Criteria

• select_context < 50ms on 10k node graph
• select_context < 200ms on 100k node graph
• Benchmark suite in tests/
• HNSW index considered for >50k nodes (may defer)

Effort: 1-2 days

Priority: Low

See: docs/proposals/memory-agent/04-implementation-roadmap.md (Phase 5)