[docs] training on specific hardware

[stevhliu]

updates the Hardware section of the docs for training:

• combined CPU/Distributed CPU into a single doc
• add more info to the Gaudi doc (mixed precision, torch.compile, distributed training)
• add more info to the MPS doc (mixed precision, model loading + device selection)
• remove the GPU doc since all that info is covered elsewhere now making it redundant

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[stevhliu]

@regisss, would you mind taking a look here please? 🙏

[pcuenca]

Took a quick look at the mps section, it looks good. Happy to take a look at the rest if you need it @stevhliu!

[stevhliu]

thanks @pcuenca! happy to get your feedback on the rest if you don't mind/have the time!