Some fixes to vec256_bfloat16.h (#59957)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/59957
Test Plan: Sandcastle
Reviewed By: VitalyFedyunin
Differential Revision: D29073913
fbshipit-source-id: dc01a2015e4ff42daa1d69443460182744c06e90
