lshift and rshift on CUDA should match the behavior on CPU (#35339)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35339
CPU version converts integers to their unsigned version first. The CUDA
version should also do it.
Also added tests for this.
Test Plan: Imported from OSS
Differential Revision: D20826862
Pulled By: ngimel
fbshipit-source-id: 164c84cfd931d8c57177a038c1bb8b6f73134d07