[FEATURE] Saving the 'epoch' argument in schedulers

You have developed great schedulers in timm. One thing that makes them a bit challenging to integrate with usual pytorch pipelines and lightning is the 'epoch' argument in the 'step' method. Pytorch saves the last step as an internal argument so calling the 'step' will automatically go one step further.

I understand that there are so many use cases that you might want to resume training from a certain start point. So, I suggest having an internal record of the epoch argument, and if the user wants to override it that can pass their intended epoch number.

If you want I can work on a PR.

The issue names no files or tests; start by locating the scheduler step methods and their existing resume or state-handling logic. Compare how the epoch argument is currently used, then verify that an internally saved epoch advances automatically while an explicitly supplied epoch overrides it.

Written by the indexing model from the issue text.