Hoist numel out of loop condtion in op_embedding (#4146)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/4146

This seems to save a single instruction in the inner loop (on x86, but I expect other architectures to be similar).
ghstack-source-id: 235302150

Reviewed By: tarun292

Differential Revision: D59335729

fbshipit-source-id: cf22669ffd8b127e60d863e4bc7858f994d8b1ce