Vec256<bfloat16>::arange step size should accept templates. (#35842)
Summary:
See https://github.com/pytorch/pytorch/issues/34555
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35842
Differential Revision: D20824448
Pulled By: ngimel
fbshipit-source-id: d76d0d499cfd102386931e4c029829e02a657bce
