[PyTorch] Migrate TI usage in ATen/native/quantized to borrowing (#58307)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58307
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: 129598791
Test Plan: Existing CI
Reviewed By: ezyang
Differential Revision: D28445922
fbshipit-source-id: ce12743980296bab72a0cb83a8baff0bb6d80091
