feat: implement stage 1 candidate retrieval and scoring logic using Qdrant and multi-factor heuristics

feat: implement matching router and stage 2 neural reranking logic

fix: use real candidate count, remove stage2 cap, add full error logging

fix: replace backend RRF with true percentage math so Stage2 final score aligns with frontend weights

feat: user-controlled pipeline depth sliders for stage1/stage2 retrieval caps

fix: cap reranker input to top-40 and fix nginx send timeout to prevent 504

feat: implement backend matching pipeline and update project documentation

feat: implement backend ingestion pipeline with Celery workers, Qdrant integration, and session management

feat: implement backend candidate matching pipeline with Pinecone integration, FastAPI routers, and ingestion workers.

feat: implement automated ingestion pipeline with state persistence and add database maintenance utilities

feat: stabilize celery loops, add redis caching, and scale stage 2 neural limit to 250

feat: add pipeline page, session delete with Qdrant cleanup, explain worker, and final frontend polish

fix: resolve asyncpg ssl connection errors and Qdrant strict payload filtering

feat: add gap analysis, LLM explainer via Groq, and all API routes

feat: implement Stage 2 reranker with BGE cross-encoder and RRF fusion

feat: implement Stage 1 retriever - ANN search + structured weighted scorer