[single_controller][decorator] Define a DynamicEnum class to make Dispatch and Execute extensible.

[hongpeng-guo]

Checklist Before Starting

• Search for similar PR(s).

What does this PR do?

Today, extending verl in proprietary usage large requires forking it, and padding code changes in the private fork.
For example, the current verl API doesn't support adding new Dispatch and Execute mode in runtime. The only way to achieve it is to make a new private fork.

This PR replace the static Enum type of Dispatch and Execute into a new "DynamicEnum" type, that the users can use new APIs register_dispatch_mode and update_dispatch_mode to adding and define new distributed mode at runtime using native verl API, instead of making a fork.

Specific Changes

• Defined DynamicEnum class in utils.py_functional.py;
• Re-defined Dispatch and Execute classes, all existing Enum API and usage are still usbale;
• Added register_dispatch_mode and update_dispatch_mode for users to register new dispatch modes at runtime;
• nit: pre-commit automatically fixed part of code format in another PR fix: ray worker exit with SYSTEM_ERROR caused by SIGALRM from math re… #1331

Usage Example

Provide usage example(s) for easier usage.

def test_register_new_dispatch_mode():
    # Test registration
    def dummy_dispatch(worker_group, *args, **kwargs):
        return args, kwargs

    def dummy_collect(worker_group, output):
        return output

    register_dispatch_mode("TEST_MODE", dummy_dispatch, dummy_collect)

    # Verify enum extension
    _check_dispatch_mode(Dispatch.TEST_MODE)

    # Verify registry update
    assert get_predefined_dispatch_fn(Dispatch.TEST_MODE) == {"dispatch_fn": dummy_dispatch, "collect_fn": dummy_collect}

Test

Added tests/verl/test_decorator.py

Checklist Before Submitting

• Read the Contribute Guide.
• Apply pre-commit checks.
• Add [BREAKING] to the PR title if it breaks any API.
• Update the documentation about your changes in the docs.
• Add CI test(s) if neccessary.