Fix zero expert gradients in bfloat16 with expert parallelism and ep_size < fsdp_size (#49688)
* Fix zero expert gradients in bfloat16 with expert parallelism and ep_size < fsdp_size
* Rewrite the expert gradient reduction comments
* Restore the expert gradient divide factor comment
* Compare the loss and every gradient in the bf16 expert parallel test
---------
Co-authored-by: Ferdinand Mom <47445085+3outeille@users.noreply.github.com>