feat(knowledge): memory graph semantic search — query processor and hybrid scoring (#3384)

[mrveiss]

Closes #3384

Implements Phase 1 (core infrastructure) and Phase 2 (hybrid scoring) from docs/design/MEMORY_GRAPH_SEMANTIC_SEARCH.md.

Summary

• query_processor.py — MemoryGraphQueryProcessor: natural language query detection, intent extraction (entity type filters, time range, status), Redis FT.SEARCH query builder, result parsing and ranking
• hybrid_scorer.py — HybridScorer: cosine similarity (embedding-based), BM25 text scoring, configurable weight blend (semantic_weight + keyword_weight), graceful degradation to keyword-only when embeddings unavailable
• __init__.py — Package re-exports public API

Tests

58 unit tests, all passing (query_processor_test.py). Redis and embeddings fully mocked.

Design decisions

• Hybrid score = semantic_weight × cosine_sim + keyword_weight × bm25 (weights configurable, default 0.7/0.3)
• Natural language detection via heuristics (verb presence, question words, sentence length) — no LLM call required
• Graceful degradation: missing embeddings fall back to keyword-only scoring without errors

Out of scope (Phases 3–5)

Context integration, caching, query suggestions, and search analytics will be addressed in follow-up issues.

[mrveiss]

Review fixes applied ✓

• Critical — HybridScorer now accepts optional redis_client in __init__; client cached on self._redis and reused across all candidates in score_and_rank — eliminates N connections per call
• Critical — _ENTITY_TYPE_PATTERNS updated to lowercase schema values (bug_fix, feature, decision, task, conversation) matching ENTITY_TYPES frozenset in schema.py
• Critical — get_related_entities now resolves entity name → UUID via get_entity_by_name() then uses memory:relations:out:{uuid} key and reads via redis.json().get() (matching graph_store schema)
• Important — hardcoded model_name="nomic-embed-text" replaced with config.get("knowledge.embedding_model", "nomic-embed-text")
• Important — all datetime.now() calls replaced with datetime.now(tz=timezone.utc)
• Critical — __init__.py unified to export both query processor/scorer symbols and graph_store/schema symbols (with try/except ImportError guard)
• Tests updated: 3 entity-type assertions updated to lowercase; get_related_entities tests rewritten to mock get_entity_by_name + redis.json().get()

All 58 tests passing.

[mrveiss]

Superseded by #3616 which consolidates all memory graph code into autobot_memory_graph.