Fix autocast for MPS devices in music generation pipeline

[Copilot]

PyTorch's torch.autocast() doesn't support the 'mps' device type, causing generation to fail on Apple Metal with RuntimeError: User specified an unsupported autocast device_type 'mps'.

Changes

• Added get_autocast_context() helper that returns nullcontext() for MPS devices (no-op since MPS pipelines use float32) and torch.autocast() for supported devices (cuda/cpu/xpu)
• Updated generation loops to use the helper instead of direct torch.autocast() calls

# Before
with torch.autocast(device_type=pipeline.mula_device.type, dtype=pipeline.mula_dtype):
    curr_token = pipeline.mula.generate_frame(...)

# After  
with get_autocast_context(pipeline.mula_device.type, pipeline.mula_dtype):
    curr_token = pipeline.mula.generate_frame(...)

The MPS path becomes a no-op context manager while CUDA/CPU behavior remains unchanged.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Generation Failed: User specified an unsupported autocast device_type 'mps'</issue_title>
<issue_description>```
Generation Failed: User specified an unsupported autocast device_type 'mps'

This is a blocker. We need to test generation in CI and make sure it actually runs on Apple MPS CPU/GPU
Our app is designed for Apple Metal CPU/GPUs and must fully support them for generation. Lets review the code carefully and fix this.</issue_description>

## Comments on the Issue (you are @copilot in this section)

<comments>
</comments>

• Fixes Generation Failed: User specified an unsupported autocast device_type 'mps' #11

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.