[te] Use c10::ScalarType utility functions in te::Dtype (#49148)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/49148
Instead of defining our own variants. I'm pretty sure this fixes a bug too, in that Bfloat16 wasn't being considered FP. Otoh, I don't think it's possible to create TEs with Bfloat so...
ghstack-source-id: 118415314
Test Plan: `buck test //caffe2/test:jit`
Reviewed By: robieta
Differential Revision: D25456767
fbshipit-source-id: bd5822114b76c4fde82f566308909bd2a55f4f21
