Add dummy bfloat16 VSX implementation (#67331)
Summary:
Just a copy of DEFAULT bfloat16 implementation and revert restriction
introduced by https://github.com/pytorch/pytorch/pull/61630
Fixes https://github.com/pytorch/pytorch/issues/66867 and https://github.com/pytorch/pytorch/issues/62016
Pull Request resolved: https://github.com/pytorch/pytorch/pull/67331
Reviewed By: ngimel
Differential Revision: D31959916
Pulled By: malfet
fbshipit-source-id: 8ca5e65ca041fef67ee18ddbb215cff01fd1e004
