🐛 Bug description

The typing for the metric_names parameter in OutputHandler-derived classes is too restrictive in several files. The signature currently expects Optional[List[str]], but according to usage in documentation and code, it should also allow a str value (e.g., "all").

• The base class in ignite/handlers/base_logger. py BaseOutputHandler is correct:

  metric_names: Optional[Union[str, List[str]]] = None

• But all the logger OutputHandler implementations in these files override it with the more restrictive type:

  • ignite/handlers/tensorboard_logger.py
  • ignite/handlers/wandb_logger.py
  • ignite/handlers/mlflow_logger.py
  • ignite/handlers/polyaxon_logger. py
  • ignite/handlers/visdom_logger.py
  • ignite/handlers/neptune_logger.py
  • ignite/handlers/clearml_logger.py

  All of these should also use Optional[Union[str, List[str]]] for metric_names for full typing correctness and consistency with the base class and library usage.

Current signature:

metric_names: Optional[List[str]] = None

Expected signature (type hint):

metric_names:  Union[str, Optional[List[str]]] = None
# or, preferably, to match the base:
metric_names:  Optional[Union[str, List[str]]] = None

This would align the typing with examples and usage where a string like "all" is passed.

Location: All logger OutputHandler classes (see list above), parameter metric_names

Steps to reproduce:

• Try type checking with metric_names="all" for OutputHandler in any logger module; type checker (e.g. mypy) issues an error.
• Documentation also shows examples where a string is allowed.

Expected behavior:

• Typing should reflect that both string (e.g., "all") and List[str] are allowed for metric_names in all OutputHandler classes.

Pull request
If this change is approved, i would like to create a PR.

Environment

• PyTorch Version:
• Ignite Version:
• OS:
• How you installed Ignite (conda, pip, source):
• Python version:
• Any other relevant information: