[Feature] Phase 5: AI/Search module (pgvector, Neo4j, RAG, MCP)

[PenguinzTech]

pgvector ai_embeddings + HNSW; embedding pipeline on Redis Streams (elder:ai:index) with markdown-aware chunking; pluggable providers (Ollama+Gemma default, OpenAI-compatible, WaddleAI enterprise); Neo4j synced projection of entities/dependencies (Cypher templates, networkx fallback); REST /api/v1/ai/{search,graph/query,rag/retrieve}; MCP tools (semantic_search, rag_retrieve, impact_analysis); AI relationship auto-recommend with approval queue; capability probing + graceful degradation. Relates to #91.

Spec: docs/design/2026-07-07-elder-platform-merge-design.md (release/v4.0.X)