feat: 10 free upgrades — streaming upload, cache invalidation, hybrid search, ONNX reranker, SQLAlchemy pool tuning, batch embed cache, semantic chunking, Prometheus histograms, DLQ auto-retry, structured logging+OTel

fix: use pdfplumber text for chunking to ensure accurate page numbers for citations

fix: keep MarkItDown for chunking, pass pdfplumber page texts separately so page numbers align for bbox extraction

fix: swap reranker to BAAI/bge-reranker-base (v2-mini does not exist on HF)

perf: replace reranker v2-m3 (1.1B) with v2-mini (100M) for 10x faster reranking

perf: 8 workers, 10 concurrent uploads, Qdrant wait=False + batch 2000

perf: revert to all-MiniLM-L6-v2, query embed cache, fire-and-forget Cloudinary, pre-warm reranker, reduce top-k

perf: upgrade embedding to all-mpnet-base-v2, add cross-encoder reranker, LRU cache, enable PII by default

production hardening: circuit breakers, graceful degradation, health checks, parallel uploads, HF Spaces deploy config

fix: search() replaced by query_points() in qdrant-client v1.9+

fix: auto-fallback to Qdrant :memory: when remote unavailable

fix: upload pipeline completely dead - worker never started, processor never registered

fix: 15 critical/high bugs across codebase

fix upload pipeline: await add_document, pre-load model eagerly, add verbose UPLOAD/VectorStore logging; remove unused deployment configs

fix models.py: match Neon schema varchar sizes, add email unique constraint