__getitem__: Ensure Tensor subclasses are not treated as tuples (#67202)
Summary:
Fixes https://github.com/pytorch/pytorch/issues/67027
`torch.Tensor` is considered a Mapping, but not a Sequence in Python
because it uses `tp_as_mapping` instead of defining `__getitem__` in
Python. However, If you try to overwrite `__getitem__` from Python
it is considered a `Sequence` and so the tensor is treated like a
tuple for indexing purposes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/67202
Reviewed By: VitalyFedyunin
Differential Revision: D31908515
Pulled By: albanD
fbshipit-source-id: 0ca55a36be3421f96428a8eacf5d195646252b38
