[PyTorch] MHA: simplify gemm_nt (#72460)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72460
Just call existing matmul (which, IIUC, handles batching itself) rather than doing a few view ops. (Please let me know if this is actually a bad idea and why!)
ghstack-source-id: 149067333
Test Plan: CI
Reviewed By: ngimel
Differential Revision: D33961843
fbshipit-source-id: ace37ad3110e1134db6c8b638ae302f0d556e00a
(cherry picked from commit 258231c0f951bd701da179eaedc1ef795416c53f)
