perf: replace local CPU inference with HF Inference API

perf: reduce per-query latency — shared model, single encode, float32 LLM

fix: resolve HfFolder import error by pinning dependencies and update README links

feat: Add reset button and optimize LLM inference speed/accuracy

feat: Complete RetailMind overhaul — hybrid retrieval, EWMA drift detection, self-healing adapter, premium UI, tests & CI