[FEAT] chore(llm): add automatic retries and fallback models

[MikeNovikoff]

Feature description

OpenRouter free tier or external APIs can be flaky. We need to handle network blips gracefully so the gRPC connection doesn't drop

What is needed to be done

• Install tenacity
• Add @Retry decorator to the LLM call with exponential backoff (e.g., max 3 attempts)
• (Optional) Implement model fallback: if openrouter/free fails, try deepseek/deepseek-v4-flash

Acceptance Criteria

• Failed API calls automatically retry without crashing the app