Update CLI docs and use mps rather than mps_device

[muellerzr]

We have the following naming convention for device vs training schema:

Device:

• Use device name, such as --tpu, --multi_gpu, --cpu

Training Schema

• Use schema name with "use" in the front, such as --use_megatron_lm and --use_deepspeed

Since MPS is a torch device rather than a schema, this PR changes the flag to be --mps and adds in a future warning for --use_mps so we can be consistent 🤗

Also updates the CLI docs to reflect the latest refactorization.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[sgugger]

Thanks for the clean up!

[pacman100]

Thanks! LGTM 🤗