Correct `Categorical` docstring (#45804)
Summary:
Clarified that the `Categorical` distribution will actually accept input of any arbitrary tensor shape, not just 1D and 2D tensors.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/45804
Reviewed By: dzhulgakov
Differential Revision: D24125415
Pulled By: VitalyFedyunin
fbshipit-source-id: 5fa1f07911bd85e172199b28d79763428db3a0f4