Update HF router models and inference flow

Replace unavailable models with HF Inference API-verified models (Mistral, Llama 2, Zephyr, etc.)

Remove Groq models; switch registry and clients to HF-only curated list

Add 15+ free HF models, expand model selection, add debug logging

Keep both Groq + HF models, remove rate-limited Groq models, unify model selection UI without backend labels

inital build