fix: docker issue

feat: migrate inference engine to Ollama for 10x faster CPU inference

fix: imoprve llm propmts

fix: default to no llm mode

chore: add Docker support for HuggingFace Spaces