feat: add Ollama chat provider

[mhsokella]

Why

Coucou's built-in chat was tied to Anthropic, which prevented users who run local models from using the file-question and chat surfaces without sending requests to a hosted provider.

What changed

• Added a selectable Anthropic or Ollama provider on Windows/Tauri and macOS.
• Added configurable model and Ollama endpoint settings, defaulting to http://127.0.0.1:11434/api/chat.
• Added Ollama request/response parsing and bounded provider error handling, while preserving Anthropic support.
• Added Rust parsing and error-handling tests, plus Windows settings/state wiring.
• Documented that Claude Code hooks observe and relay an already-running Claude Code process; selecting Ollama does not drive or replace Claude Code. The existing fail-open hook behavior remains documented.

Validation

• Windows TypeScript type-check passes with npx tsc --noEmit.
• git diff --check passes.
• Rust tests/build could not run because Cargo is unavailable in this Windows environment.
• macOS build could not run because xcodebuild is unavailable in this Windows environment.

Notes

Ollama chat is text-only in this implementation, so non-text file context is not sent as image/PDF content to the local provider.

[Louis-CFM]

Thanks a lot for this. Local models landed in #156, which adds Ollama and LM Studio to the Mac chat with streaming and a connect flow, so I am closing this one. Thanks again for pushing the idea.