Add RayClusterHandler to enable fairseq2 run on Ray

[dongwang218]

What does this PR do? Please describe:
Allow fairseq2 to run inside ray cluster.

Does your PR introduce any breaking changes? If yes, please list them:
List of all backwards-incompatible changes.

Check list:

• [N ] Was the content of this PR discussed and approved via a GitHub issue? (no need for typos or documentation improvements)
• [ Y] Did you read the contributor guideline?
• [ Y] Did you make sure that your PR does only one thing instead of bundling different changes together?
• [ Y] Did you make sure to update the documentation with your changes? (if necessary)
• [ Y] Did you write any new necessary tests?
• [ Y] Did you verify new and existing tests pass locally with your changes?
• [ N] Did you update the CHANGELOG? (no need for typos, documentation, or minor internal changes)

[artemru]

nice contribution !
can you add a simple integration test and some doc so everyone could start using this integration with Ray ?

[cbalioglu]

I believe this should be optional. In fact, instead of making it optional, we can just do a quick import check in the cluster implementation, and raise an error advising the use to install Ray themselves. For majority of use cases, people won't be using Ray. See this as an example.

[dongwang218]

made ray as extra. also make sure the test and existing code works without install ray.

[cbalioglu]

You can (ideally should) use the get_rank, get_local_rank, get_world_size, and get_local_world_size utility functions defined here: https://github.com/facebookresearch/fairseq2/blob/main/src/fairseq2/utils/env.py#L85

They also raise an appropriate exception that causes the process to terminate gracefully.

[dongwang218]

thanks, changed to use utility functions.

[artemru]

I would be interested in reanimating this work!

[dongwang218]

I would be interested in reanimating this work!

sure, will merge master and push the latest.