fix(hf): upgrade llama-cpp-python 0.3.10, warm-up Ollama, Streamlit deprecation

fix: use Sekoya fine-tuned model + mmproj download for HF Spaces

feat: complete DDD architecture and HF deployment