Add an experimental support to LLMs in MLX format as a standalone target (SwiftAIMLX)

Features:

• ✅ Text generation
• ✅Tool use (if the model supports it)
• ❌Structured Outputs (currently not supported)

MLX model weights can be downloaded from Hugging Face.

• Also include a demo app
• Changes are reflected in 001-llm-api.md

Enhance API Safety and Type Inference

• Define errors out in several API: Chat.init, LLM.makeConversationThread, ...
• Make the Message API a sealed enum instead of protocol for better safety
• Better type inference for @Guide

SwiftAI is a library for building LLM powered iOS and macOS applications. This initial alpha release provides a unified API that works seamlessly across Apple's on-device model and OpenAI's LLMs.

Key Features

• Model Agnostic API
• Typesafe Structured Output
• Tool Use