[PyTorch] Use .sizes() instead of .size() in _cat_out_cpu (#49368)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/49368
The former is faster because it doesn't allow negative indexing (which we don't use).
ghstack-source-id: 118624598
Test Plan: internal benchmark
Reviewed By: hlu1
Differential Revision: D25545777
fbshipit-source-id: b2714fac95c801fd735fac25b238b4a79b012993
