LLM-backed chat via HF Inference (Gemma 3 27B) with fixed rule-based fallback

Fix PDF report cursor bug; add robust file loading and richer preprocessing options

Initial deploy: ML Data Analysis Studio