Enable `ceil`, `floor`, `frac`, `round` & `trunc` for BFloat16 on CUDA (#57910)
Summary:
Enable `ceil`, `floor`, `frac`, `round` & `trunc` for BFloat16 on CUDA
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57910
Reviewed By: soulitzer
Differential Revision: D28579486
Pulled By: ngimel
fbshipit-source-id: 2f90354339dbccb69cea7ec9caf9b066ea13a666