Remove type subclassing (#24257)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/24257
Type subclassing was used to support our old hierarchy of
Tensor types. Now that we have one tensor type it is not needed.
This removes:
* isSubclass, since it is now always false.
* type slicing, which was only needed for subclasses.
* AutogradZeroTensor, which is folded into ProfiledTensorType
Test Plan: Imported from OSS
Reviewed By: jamesr66a
Differential Revision: D16794035
Pulled By: zdevito
fbshipit-source-id: 9a3e6101df0d51029a5e667a9c9137d2ae119aa7