Agent V3 Memory Specialist

This consolidates seven different memory systems (MemoryManager, SwarmMemory, SQLiteBackend, etc.) into a single AgentDB backend with HNSW vector indexing. The promised speedup is 150x to 12,500x on semantic searches, which matters if you're dealing with large memory datasets across multiple agents. It's part of the V3 architecture overhaul implementing ADR-006 and ADR-009. The migration strategy is gradual with backward compatibility, so you're not forced to rip everything out at once. Most interesting for multi-agent setups where cross-agent memory sharing and fast retrieval actually matter, less so if you're running simple single-agent workflows.