feat: add OpenRouter as cloud LLM provider

Problem

MPV2 currently only supports Ollama for LLM text generation. This requires running a local Ollama server, which is a setup barrier for many users.

On #139, you mentioned it would be smarter to add OpenRouter instead of individual providers.

Proposed Solution

Add OpenRouter as an alternative LLM provider. OpenRouter exposes an OpenAI-compatible API that routes to 200+ models (GPT-4o, Claude, Gemini, Llama, Mistral, etc.) through a single API key.

Configuration:

{
  "llm_provider": "openrouter",
  "openrouter_api_key": "your-key",
  "openrouter_model": "openai/gpt-4o-mini"
}

Ollama stays the default. No breaking changes. I have an implementation ready that follows the existing llm_provider.py patterns.