Fix TTS Support For Apple MPS

[zboyles]

Summary

This pull request adds MPS (Apple Silicon) device support for TTS with a non-breaking change to the Mimi model weight loading in src/liquid_audio/processor.py.

Problem

The safetensors library doesn't support loading weights directly to MPS devices, causing an error: device mps:0 is invalid when attempting to use the library on Apple Silicon Macs.

Solution

Added device detection logic in the mimi property to:

• Load weights to CPU first when the target device is MPS
• Let PyTorch automatically transfer weights to MPS during load_state_dict()
• Preserve existing behavior for CUDA, CPU devices, and MPS devices using CPU (no changes)

Changes

• Modified src/liquid_audio/processor.py: Added MPS device detection to conditionally load weights via CPU as an intermediate step

Impact

• ✅ Non-breaking: No changes to existing CUDA/CPU workflows
• ✅ Enables MPS support: Users can now run on Apple Silicon devices
• ✅ Minimal change: Single-line conditional added for device compatibility

[Copilot]

Pull Request Overview

This pull request fixes TTS support for Apple Silicon Macs by addressing a safetensors library limitation with MPS device loading. The change enables MPS device compatibility while maintaining existing behavior for CUDA and CPU devices.

• Adds device detection logic to handle MPS devices when loading Mimi model weights
• Implements CPU-first loading strategy for MPS devices as a workaround for safetensors limitations
• Preserves existing behavior for all other device types

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}