Add pytorch lr scheduler

[DNA386]

This PR adds the option of including a learning rate scheduler to the training loop.
Requires adding an extra method called after each epoch is completed. By default do nothing to maintain previous functionality.
By default no schedule is set.

[neiljdo]

@DNA386 thank you for this PR. I believe this only works with epoch-based schedulers and not with other types (e.g. step-based like OneCycleLR, etc.)?

[DNA386]

Ah you're right, I didn't see there was a different kind.
The cleanest way to unify this and also allow chaining multuple together might then be to add a new scheduler lambeq class, with subclass pytorchScheduler that can wrap (possibly multiple) pytorch schedulers.
scheduler class api will then get a position property (epoch vs step) and generic call function that takes in the current epoch, loss, and step (and whether this is the last step in this epoch). In this way can then potentially revert the code back and call the scheduler after each step.

I will implement and update the PR