Most of our prototype transforms that override forward do this only to perform some input checks before passing them on to _get_params and _transform. This PR adds a _check method that makes this regular behavior, whereas overriding forward means some "special" behavior the API can't handle.

with _check in place, there are only three* transforms that still need to override forward:

• vision/torchvision/prototype/transforms/_augment.py

  Line 188 in 7cc2c95

  class SimpleCopyPaste(_RandomApplyTransform):

• vision/torchvision/prototype/transforms/_auto_augment.py

  Line 139 in 7cc2c95

  class AutoAugment(_AutoAugmentBase):

• vision/torchvision/prototype/transforms/_container.py

  Line 10 in 7cc2c95

  class Compose(Transform):

These are objectively outlier and thus it is justified for them to override forward.

Using _check will also get rid of the boilerplate

def forward(self, *inputs):
    sample = inputs if len(inputs) > 1 else inputs[0]

    ...

    return super().forward(sample)

idiom.