Update LLMEval example

[timsneath]

Proposed changes

This PR modernizes the LLMEval example application by refactoring it from a monolithic ContentView into a clean MVVM architecture with several new features.

New Features

• Improved metrics panel: Better visual hierarchy for performance statistics
• Preset prompts: Curated prompt library with support for tools and thinking mode, including long-form prompts
• Enhanced loading UX: Visual progress indicators for model downloads with file counts
• Collapsible prompt area: Expandable text input for longer prompts

Architecture Refactor

• Extracted business logic into LLMEvaluator view model with improved state management
• Split UI into focused, reusable components: HeaderView, OutputView, PromptInputView, MetricsView,
  LoadingOverlayView, PresetPromptsSheet
• Created dedicated service layer with ToolExecutor for function calling
• Organized models into separate files (PresetPrompts, ToolDefinitions)

Configuration Updates

• Changed default model from Phi-2 to Qwen3-8B-4bit to demonstrate improved prompt loading on M5 hardware
• Updated README with clearer instructions for switching models
• Added an app icon
• Cleaned up entitlements

Checklist

Put an x in the boxes that apply.

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

[davidkoski]

The changes look good, but some things to think about:

• I think we want a very simple example that people could start with -- LLMEval is probably that app

  • that isn't to say that it shouldn't be improved or cleaned up, but feature-wise I think it should stay pretty bare
  • and ideally with few lines of code
  • but it hasn't been touched much for a long time so it can probably be modernized a bit!

• the MLXChatExample app, on the other hand, is a little more feature rich and shows what you might build

  • I wonder if these changes might be more appropriate there?

OR

• do we need a new application that better showcases features like feeding it the .md files as part of the prompt?
  • then we could actually simplify the LLMEval app to contain just the barest features

We should probably provide more guidance documentation-wise as to what to expect from each.

What do you think?

[davidkoski]

After offline discussion the new plan is:

• take these improvements
• I will make a new minimal "hello world" LLM application as a minimal example

[davidkoski]

Thank you for this improvement!