Fix arange functions for VSX specializations of Vec256 (#58553)
Summary:
Need a templated 2nd parameter to support e.g. double steps even for int vectors.
This extends https://github.com/pytorch/pytorch/pull/34555 x86 specific fix to VSX instruction set.
Fixes https://github.com/pytorch/pytorch/issues/58551
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58553
Reviewed By: ailzhang
Differential Revision: D28551266
Pulled By: malfet
fbshipit-source-id: de7d23685da06b1b3089933d74398667cfb43c9f
