Mac Support: Adapt FramePack for Apple Silicon

[donghao1393]

This PR adds support for running FramePack on Apple Silicon Macs using MPS (Metal Performance Shaders) backend.

Changes

• Added basic MPS support tests to verify functionality on Apple Silicon
• Added MPS attention mechanism tests to verify the chunked attention implementation
• Added documentation for Mac users
• Added MPS fallback for unsupported operations
• Based on PR Tinygrad MPS lllyasviel/FramePack#65 by @mdaiter, which implemented the core MPS adaptations

Testing

• Tested on M4 Max with 128GB memory
• Basic MPS support tests pass successfully
• MPS attention mechanism tests pass successfully

Documentation

• Added with instructions for Mac users
• Included performance considerations and troubleshooting tips
• Added instructions for handling unsupported MPS operations

Usage

For operations not yet implemented in MPS, we use the environment variable to fall back to CPU. This is handled automatically in the code and documented in the troubleshooting section.

A convenience script is provided to run the demo with the correct environment variables set.

This PR is a work in progress and will be updated with additional optimizations as needed.

[mdaiter]

(Thank you @donghao1393 ! Can send yo more as I keep digging)