[PyTorch] Remove aten::native::empty usage in TensorIndexing (#49074)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/49074
Try to resolve part of the github issue of https://github.com/pytorch/pytorch/issues/48684 . ```aten::native::empty()``` is referenced in TensorIndexing.h. However, the definition of ```aten::native::empty()``` is nothing but checks and eventually calling ```at::empty()```.
In this diff, ```at::empty()``` is directly used to avoid the reference to native symbols.
ghstack-source-id: 118165999
Test Plan: CI
Reviewed By: dhruvbird
Differential Revision: D25417854
fbshipit-source-id: 7e4af411ae63642c8470e78cf8553400dc9a16c9
