Add "Try a local model" one-click Ollama setup for out-of-the-box AI

[faratech]

Type: Enhancement (feature request from user feedback — not a bug)

Motivation: Most users won't paste an API key. From the thread: "lets face it most people aren't going to paste an API key… they are going to want it to 'just work'. It's not advanced users that need this, it's the masses." AI features should work out of the box for non-technical users.

Proposal: Add a "Want to try a local model?" call-to-action (e.g. on the AI screen empty state and/or in Settings → Provider setup) that, in one click:

1. Installs Ollama if it isn't present (winget install Ollama.Ollama, which the app already references elsewhere as the Foundry/Ollama path), and
2. Pulls a small, fast model suitable for this hardware (e.g. a ~1–3B model), then
3. Sets preferredAIProvider + ollamaModel so the assistant is immediately usable.

The provider plumbing already exists — ai_providers/ollama.rs supports Ollama, endpoint auto-discovery, and model resolution — so this is primarily an onboarding/install-orchestration flow with progress UI (download can be large, so show progress and let the user cancel).

Considerations:

• Surface size/time expectations before downloading (Ollama + a model is hundreds of MB to a few GB).
• Gracefully handle no-NPU / low-RAM machines (pick a small model, or fall back to a clear message).
• Keep it optional and reversible.

Related: the AI provider selector clarity work (#25) and the local-model settings UX (#24) — once one-click setup exists, those screens should reflect the now-configured local model.