mlflow run context is not logged when using MLFlowLogger

[stllfe]

🐛 Bug

When we use the basic mlflow logging via with mlflow.start_run(): ... context manager, we get a better supplementary info about the run (git commit sha, user, filename) rendered in the Tracking UI (system tags)

But when we use MLFlowLogger as a logger in pytorch_lightning, this info is not logged. As a user, I'd like to have a mirrored functionality out-of-the-box.

I inspected the start_run() method of mlflow and deduced that the only thing is left while creating the run via MLflowClient is to add resolve_tags from the context package:

# pytorch_lightning/loggers/mlflow.py
from mlflow.tracking.context.registry import resolve_tags
...
    def experiment(self) -> MLflowClient:
        if self._run_id is None:
            run = self._mlflow_client.create_run(experiment_id=self._experiment_id, tags=resolve_tags(self.tags))

I think it's a better idea to add those tags internally (meaning not to expect users doing that manually) as first - it's as seamless as in the default API, secondly - it's the pytorch_lightning that manages the mlflow's run anyways.

PR is following ...