[PyTorch] Migrate TI usage in ATen/native/cpu to borrowing (#58303)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58303
Borrowing is more efficient, and we can see in all these cases that the TensorIterator doesn't outlive the input & output Tensors.
ghstack-source-id: 129471191
Test Plan: Existing CI
Reviewed By: ezyang
Differential Revision: D28444032
fbshipit-source-id: f6a9e9effb43c273f464ef6ff410274962f3ab23
