feat: Allow using local Ollama models for AI assistant

[BinaryMuse]

This PR lays the groundwork for allowing model selection for the AI features; currently, only local Ollama is supported (in addition to Atuin Hub); others may be added following similar patterns.

On hold pending jeremychone/rust-genai#124 (updated fork instead)

[greptile-apps]

Greptile Summary

Adds support for using local Ollama models as an alternative to Atuin Hub for AI features. Changes span both backend (Rust) and frontend (TypeScript/React).

Key changes:

• Backend now accepts model selection per message and supports no-auth providers (Ollama)
• Tool results now use individual ChatRole::Tool messages instead of batching
• New settings UI for provider selection and Ollama configuration
• Model selection validated before sending messages

Issues found:

• URL joining logic has bug with protocol handling (http:// becomes http:/)
• Missing validation for empty model name when Ollama is enabled

Important Files Changed

Filename	Overview
backend/src/ai/session.rs	Added model parameter to send_user_message, changed API key to Option to support no-auth providers
backend/src/ai/fsm.rs	Changed tool results to use individual ChatRole::Tool messages instead of batching in user message
src/state/settings_ai.ts	New file: manages AI provider settings and model selection logic
src/components/runbooks/editor/ui/AIAssistant.tsx	Retrieves selected model from settings and validates configuration before sending messages

[greptile-apps]

_{11 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}