[docs] Clarify batch_first behavior for nn.LSTM, nn.RNN, and nn.GRU (#58809)
Summary:
Fixes the high-pri doc component of https://github.com/pytorch/pytorch/issues/4145.
To make the input / output shapes more readable for both `batch_first` states, this PR also introduces short dim names. Opinions welcome on the readability of the restructured docs!
Screenshot for `nn.LSTM`:
<img width="791" alt="Screen Shot 2021-05-24 at 5 11 39 PM" src="https://user-images.githubusercontent.com/75754324/119408130-389e5300-bcb3-11eb-9a4f-1df96a0a4d70.png">
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58809
Reviewed By: gchanan
Differential Revision: D28685415
Pulled By: jbschlosser
fbshipit-source-id: e8c92e3d7e052071a505b55dca976fd2ef5a8307
