feat: add MemoryStore with deduplication-aware storage

[daniel-harperdb]

Summary

Adds a MemoryStore custom resource with built-in deduplication.

• Before inserting, searches for similar existing memories using vector similarity
• Configurable dedupThreshold (default 0.95) — memories above this similarity are considered duplicates
• On duplicate: updates metadata and timestamp instead of creating a new record
• On new: generates embedding server-side and inserts
• Returns { action: "created" | "deduplicated", id: string }

Stacked on #8 (score-normalization).

Test plan

• Unit tests included (test/dedup-store.test.js)
• Manual test: store same text twice, verify dedup behavior
• Manual test: store similar but different text, verify new record created
• Test with custom dedupThreshold values