Temporal transform: rearrange temporal and channel dimensions

[yichiac]

This PR is to add a temporal transform: rearrange temporal and channel dimensions, as mentioned in #2382.

[adamjstewart]

Would be better to have a single PR for each transform instead of multiple in 1 PR.

[yichiac]

Not sure if the class names and functions are good. Tagging @adamjstewart for review.

[adamjstewart]

My vote would be for a single class that simply wraps around einops.rearrange like so:

class Rearrange(GeometricAugmentationBase2D):
    def __init__(self, *args: Any, **kwargs: Any) -> None:
        super().__init__(p=1)
        self.flags = {'args': args, 'kwargs': kwargs}

    def apply_transform(
      self, input: Tensor, params: Dict[str, Tensor], flags: Dict[str, Any], transform: Optional[Tensor] = None
   ) -> Tensor:
        return rearrange(input, *flags['args'], **flags['kwargs'])

Note that GeometricAugmentationBase2D has additional functions that need to be overridden, see https://kornia.readthedocs.io/en/latest/augmentation.base.html#custom-augmentation-classes for details.