Negate halves on GPU using __hneg() when possible, instead of using float conversion.
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/23626
Test Plan: Imported from OSS
Differential Revision: D16656730
Pulled By: ezyang
fbshipit-source-id: 7e1f4e334f484a3ed4392949ff7679cefd67a74e