fix(tracking): default step=None on tracker.log and accept extra kwargs in MLflowTracker

[1fanwang]

Closes #4041.

What does this PR do?

Two related bugs in src/accelerate/tracking.py:

1. tracker.log() raises TypeError when step is omitted.

Every tracker subclass declares step: Optional[int] with no default value:

def log(self, values: dict, step: Optional[int]):
    ...

The docstrings all describe step as optional. Users following the docstring hit:

TypeError: log() missing 1 required positional argument: 'step'

Fix: set step: Optional[int] = None on GeneralTracker.log, TensorBoardTracker.log_images, AimTracker.log, MLflowTracker.log, and the MyCustomTracker test fixture (which was propagating the broken signature into test scaffolding).

2. MLflowTracker.log swallows log_kwargs.

Accelerator.log(values, log_kwargs={"mlflow": {"synchronous": True}}) is documented to forward provider-specific kwargs through to each tracker. MLflowTracker.log doesn't accept **kwargs, so the call raises:

TypeError: log() got an unexpected keyword argument 'synchronous'

Fix: accept **kwargs in MLflowTracker.log and forward to mlflow.log_metrics.

How was this tested?

Two regression tests under MLflowTrackingTest:

• test_log_without_step — tracker.log({"loss": 0.1}) with no step.
• test_log_accepts_extra_kwargs — accelerator.log({"loss": 0.1}, step=1, log_kwargs={"mlflow": {"synchronous": True}}) and assert mlflow.log_metrics was called with synchronous=True.

Before (against upstream/main):

$ pytest tests/test_tracking.py::MLflowTrackingTest::test_log_without_step -v --tb=short
tests/test_tracking.py::MLflowTrackingTest::test_log_without_step FAILED
=================================== FAILURES ===================================
___________________ MLflowTrackingTest.test_log_without_step ___________________
tests/test_tracking.py:331: in test_log_without_step
    tracker.log({"loss": 0.1})
src/accelerate/tracking.py:89: in execute_on_main_process
    return PartialState().on_main_process(function)(self, *args, **kwargs)
E   TypeError: MLflowTracker.log() missing 1 required positional argument: 'step'
1 failed in 3.95s

$ pytest tests/test_tracking.py::MLflowTrackingTest::test_log_accepts_extra_kwargs -v --tb=short
tests/test_tracking.py::MLflowTrackingTest::test_log_accepts_extra_kwargs FAILED
=================================== FAILURES ===================================
_______________ MLflowTrackingTest.test_log_accepts_extra_kwargs _______________
tests/test_tracking.py:346: in test_log_accepts_extra_kwargs
    accelerator.log({"loss": 0.1}, step=1, log_kwargs={"mlflow": {"synchronous": True}})
src/accelerate/accelerator.py:3381: in log
    tracker.log(values, step=step, **log_kwargs.get(tracker.name, {}))
src/accelerate/tracking.py:89: in execute_on_main_process
    return PartialState().on_main_process(function)(self, *args, **kwargs)
E   TypeError: MLflowTracker.log() got an unexpected keyword argument 'synchronous'
1 failed in 3.08s

After (this PR):

$ pytest tests/test_tracking.py::MLflowTrackingTest::test_log_without_step tests/test_tracking.py::MLflowTrackingTest::test_log_accepts_extra_kwargs -v
tests/test_tracking.py::MLflowTrackingTest::test_log_without_step PASSED
tests/test_tracking.py::MLflowTrackingTest::test_log_accepts_extra_kwargs PASSED
2 passed in 4.18s

Before submitting

• This PR fixes a typo or improves the docs (if so, you can skip checks)
• Did you read the contributor guideline?
• Was this discussed/approved via a Github issue or the forum? Closes MLflowTracker.log drops log_kwargs; tracker.log requires step despite docstring #4041.
• Did you make sure to update the documentation with your changes? No doc change — docstrings already declared step optional, the code now matches.
• Did you write any new necessary tests?