Migrate dirichlet_grad from CUDA_tensor_apply4 to TensorIterator (#33996)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/33996
Test Plan: Imported from OSS
Differential Revision: D20196789
Pulled By: VitalyFedyunin
fbshipit-source-id: 69ee720f4f3d8a2df91874b77ee3918ce1b951b2