Add LLM actions endpoint that generates hour-specific heat management recommendations. Replace static action engine with AI-driven approach. Add cool mode logic (ventilate/ac/overloaded), indoor temperature tracking, and timeline legend with annotations.
Add Gemini provider using the generativelanguage.googleapis.com API with systemInstruction support. Activated via --llm gemini with GEMINI_API_KEY env var. Default model: gemini-2.0-flash.
