Mostly sync BlasKernel.cpp with ATen ReducedPrecisionGemvFastPathKernel

Pull Request resolved: #10868

The two files were similar, but diverged due to recent changes. Since we have sharing of PyTorch headers, we can keep them mostly the same; differences are some of the namespace stuff and a couple of EXECUTORCH NOTEs.
ghstack-source-id: 284071626
@exported-using-ghexport

Differential Revision: [D74702689](https://our.internmc.facebook.com/intern/diff/D74702689/)

Author: swolchok