Add basic support for mps (Apple Silicon) as an accelerator

[colbyford]

This PR includes minor changes to the click options for --accelerator to allow mps as an option. This enables GPU support on local devices with Apple Silicon automatically by passing this option through the PyTorch Lightning Trainer() class.

Notes and limitations:

• In the original code, device_type is hardcoded as "cuda" here. Future updates should pass through the device information automatically to avoid warnings in the torch.autocast(...) method.
• There are warnings that parts of the process are not supported on MPS. (e.g. pin_memory is unused and aten::linalg_svd will run on CPU), though it folds successfully.
• The PR also includes a basic Dockerfile template (which is independent from the MPS support changes).

[fnachon]

I already made a PR for Apple Silicon, but it was not considered yet: #231