feat: cache reranker snapshot at build time via huggingface_hub

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

fix: skip reranker build-time cache (1.1 GB OOMs builder), lazy-load at runtime

fix: free memory between model downloads, graceful fallback if reranker OOMs

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