v2 final: llama-cpp-python in Docker + TinyLlama pre-download + auto-provider streamHF + setup endpoint

fix: HF inference SDK v3 + InferenceClient + supported model IDs (Llama 3.2, Gemma, Phi-3)

fix: use router.huggingface.co (api-inference.huggingface.co blocked from Spaces)

debug: test router.huggingface.co + groq + openrouter connectivity

debug: add /api/debug/connectivity endpoint to diagnose network issues

fix: streamHF uses raw fetch (not SDK) — better error messages + HF 401 debug

fix: single-stage Dockerfile + .dockerignore + no node-llama-cpp + llama.cpp REST sidecar

v2 final: clean Dockerfile + node-llama-cpp local model + full UI (markdown+artifacts+fileupload)

fix: simplified Dockerfile (no llama compile) + background local model download + full UI

v2: Full claude.ai clone — TinyLlama local model + markdown + artifacts + file upload + multi-provider

🚀 Major upgrade: multi-provider (Groq/Together/OpenRouter/HF), AirLLM local GPU, web search, DeepSeek V3/R1

fix: remove verbose:false from better-sqlite3 (expects function not bool)

🚀 Initial Fable AI deployment — open-source models via HF Inference API