Switch JD-parse/T1 polish to Qwen3-1.7B

Respect --limit; LLM streaming heartbeat

Add LLM call timeout and streaming pipeline phases

Add observability logging for LLM and embedder

Add timing and error logging throughout

Cache-first model loading for zero-network ranking

Add CPU threading, T1-only LLM polish, experiment script

Reuse Qwen3.5 for T2; drop LFM2.5 & factor ONNX

Explain: research-based outcome notes (prompt-engineered + verified)

Models load fully offline + enable runtime auto-download

Docker: pre-download all models at build for offline ranking + docs

Multi-band explainability: tiered NLG + dual-model polish (Step 11)

Enhance JD extraction (Qwen3.5-2B + richer prompt) and Step-11 reasoning

Add reasoning LLM, perf tweaks, and UI timing

Add JD parsing, central config, checkpoints & LLM