[Sub-Task] Re-Ranker Middleware Construction (Hybrid RAG Pipeline)

Origin Session ID: af26000d-914a-4eb0-8d28-2c09e9cb4cb5 Parent Epic: #9922

Context

This is the core functional engine of the Two-Pillar Hybrid RAG implementation. We must replace the legacy $O(N)$ execution inside SQLiteVectorManager.mjs with an intelligent, dual-pass GraphQL-style Re-Ranker.

Technical Requirements

1. Vector Deprecation: Formally disable or extract SQLiteVectorManager.mjs from actively returning distance geometry. Set ChromaManager.mjs as the binding agent for AbstractVectorManager.
2. The Re-Ranker (StorageRouter.mjs): Implement the Phase 1 / Phase 2 query pipeline constraint:
   • Pass 1: Execute HNSW similarity search via Chroma DB returning Top K broad matches.
   • Pass 2: Filter those returned Chroma IDs structurally through the Native Edge Graph (SQLite). Score candidates based on edge proximity, sequence continuity, and explicit SIMILAR_TO links.
3. Weighting Logic: Develop the basic math function to combine the mathematical distance from Chroma with the topological edge depth from SQLite to provide a unified relevanceScore array to the LLM agent.

Definitions of Done

• StorageRouter.getMemoryCollection() cleanly routes Semantic lookups to Chroma.
• Queries explicitly enforce the two-pass pipeline, validating the Node JS / C++ hybrid execution speed.