Hybrid GraphRAG Phase 2: Pipeline Tuning & Deterministic Topology

Problem

The initial iteration of the Hybrid GraphRAG pipeline successfully stood up the Native Edge Graph (SQLite) and basic Tri-Vector synthesis (Issue #9673). However, the sandman_handoff.md file suffers from infinite bloat due to unpruned appending, and the Capability Gap Inference generates false positives (Docs/Test gaps) by relying heuristically on LLM text output rather than deterministic codebase artifacts (e.g., jsdocx.mjs). By shifting to deterministic graph edges (e.g., mapping MISSING_DOCS directly from valid structural paths), we bypass hallucinated gaps and secure strategic mapping.

Sub-Tasks

• Implement Deterministic Edge Mapping for Document Gaps (docs/output/structure.json)
• Implement Deterministic Edge Mapping for Playwright Test Gaps
• Centralize sandman_handoff.md regeneration in DreamService.synthesizeGoldenPath
• Enforce TTL (Time-to-Live) Pruning for historical Topological Conflicts
• Refactor LLM ReAct Gap pass to strictly evaluate "Strategic Drift"

Definition of Done

• sandman_handoff.md correctly scales, is overwritten idempotently, and cleanly organizes the Golden Path and Structural gaps.
• executeCapabilityGapInference no longer flags adequately documented classes (e.g. MainContainer) as missing documentation.
• All gaps are tracked natively as formal SQLite Edge relationships (MISSING_DOCS, MISSING_TESTS, CAUSES_CONFLICT).