Fix backend inference logging and improvements

Improve agent chat tone and HF fallback behavior

Fix Space runtime HF router integration

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

Add contextual chat message generation when HF models don't return messages

Add comprehensive logging for HF API token and inference failures

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

Add HF fallback for Groq rate-limit errors; remove rate-limited Groq models

Remove rate-limited Groq models (llama-3.1 variants, gemma) and use Mixtral as default

Fix place-fire random range failures

Switch to dynamic model-generated chat with context

Update simulation chat personality and celebration messages

Deploy backend from Rush-Agents-Rush