MPS Mixed-precision Autocast

Nobody has claimed this yet.

### Description & Motivation

Support for MPS autocasting has recently be added in PyTorch 2.5.0 [here](https://github.com/pytorch/pytorch/pull/99272) and there is an ongoing effort to implement gradient scaling [here](https://github.com/pytorch/pytorch/issues/142397).

PyTorch Lightning does not currently support mixed-precision on MPS device but it could be added in a near future when gradient scaling is finalized.

Is this feature considered? This would allow reducing memory usage and improving training time for some models.

### Pitch

Currently PyTorch Lightning falls back to FP32 when trying to use mixed-precision and issues a warning mentioning CUDA.

I think that considering adding a path for MPS mixed-precision would be great.

### Alternatives

Stick to FP32 training when using MPS device.

### Additional context

thanks for your work!

cc @lantiga @borda @justusschock

Start by reviewing the linked PyTorch MPS autocasting pull request and the gradient-scaling issue to confirm the supported APIs and their status. Then trace PyTorch Lightning's mixed-precision device handling and its CUDA warning path; done means MPS mixed-precision selects the supported path without falling back to FP32, with coverage for the relevant behavior.

Written by the indexing model from the issue text.