Automate Gemini model lifecycle to prevent deprecation breakage

[hessius]

Problem

MeticAI hardcodes the Gemini model name (gemini-2.5-flash as of v2.2.2). When Google deprecates a model, all users on that version break with 404 errors. This just happened with gemini-2.0-flash (see #292).

Relationship to Existing Issues

This is related to #180 (Select model, milestone 2.4) which covers user-facing model selection. This issue focuses specifically on preventing silent breakage from model deprecation.

Ideas to Explore

1. Auto-updated model alias: Use gemini-flash or similar alias that Google auto-updates to latest stable. Need to verify this works reliably with the google-genai SDK.

2. Model availability health check: Add a startup or periodic check that tests whether the configured model is accessible. If not, log a warning and optionally fall back to a known-good model.

3. Fallback chain: Configure a list of models to try in order (e.g. gemini-2.5-flash → gemini-2.0-flash → gemini-flash-latest). On 404, try the next one.

4. Settings UI for model selection (ties into Select model #180): Let users pick their model from a dynamically fetched list of available models via the Gemini API's list_models() endpoint.

5. Version notification: When a new MeticAI version is available, notify users proactively (could tie into Watchtower notifications).

Acceptance Criteria

• MeticAI should not silently break when Google deprecates a model
• Users should get a clear, actionable error message if their model becomes unavailable
• Ideally, the system auto-recovers without user intervention