Add Gradio interface for LLM safety evaluation and implement inference provider cost fetching

Refactor app.py to improve dataset metadata retrieval and enhance Gradio interface with better prompt field validation and export options

Update default dataset selection in Gradio interface to 'ai-safety-institute/AgentHarm'

Enhance device management: add get_device function for CUDA detection and update model loading in app.py and eval_llm.py

Refactor LLM safety evaluation: remove app.py, update Dockerfile, enhance README, and add dataset loading script