feat(llm): implement multi-profile system for local and remote endpoints

[thedavidweng]

Summary

This PR introduces a robust LLM profile management system that allows users to configure and
switch between multiple remote and local models seamlessly. It transitions from a single global
LLM setting to a dynamic, multi-provider architecture.

Key Features

• Dynamic Profile Management: Users can now create, name, and delete multiple LLM profiles (e.g., separate profiles for OpenAI, local APFEL/Ollama, or MLX).
• APFEL Integration: Added a one-click shortcut to set up a profile for "APFEL" (local Apple Foundation Models via Ollama-compatible endpoints).
• Local MLX Support: Built-in model picker and download management for Apple Silicon-optimized local models.
• Native UI: Redesigned the "LLM" tab in the Setup Wizard with a new profile selector and controls.
• Core Refactoring:
  • Updated the Rust core to handle profile resolution at runtime via a BTreeMap based profile store.
  • Added atomic JSON-based FFI endpoints (sp_llm_profiles_json, etc.) for seamless communication with the native Objective-C shell.
  • Implemented backward-compatible migration logic for existing configurations.

Technical Details

• FFI: Added support for sp_llm_profiles_json, sp_llm_save_profiles_json, and sp_llm_test_profile_json.
• Config Schema: Refactored LlmSection to store active_profile and a collection of profiles.
• Reasoning Control: Improved support for reasoning_effort (OpenAI o-series) and thinking (GLM etc.) at the profile level.

Verification

• Verified full build with make build.
• Passed unit tests in koe-core/src/config.rs for profile resolution and default states.
• Manually verified profile switching, adding, and saving in the native macOS UI.

[missuo]