Remove hardcoded NVIDIA API key - now requires env var

Add NVIDIA API (GLM-4.7) as primary inference method with thinking mode

Add detailed logging to diagnose LLM inference failures

Clean repository root for HF Spaces