Avoid calling allclose in the backward if there are tensor subclasses (#91444)
`allclose` it's data-dependent (returns a bool) so it does not play well
with functorch. We are skipping that check in the context of subclasses
to avoid hard errors.
Partially fixes https://github.com/pytorch/pytorch/issues/90499
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91444
Approved by: https://github.com/albanD
