Merge TensorPipe's CPU and CUDA channel registry (#59375)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59375
The CPU and CUDA channels used to be separate classes in TensorPipe, but they recently got merged in order to support cross-device-type channels. We used to need two separate registries in PyTorch, but now we can merge them. This simplifies some registration logic, and will help in future PRs.
ghstack-source-id: 130583770
Test Plan: CI
Reviewed By: mrshenli
Differential Revision: D28796427
fbshipit-source-id: b7db983293cbbddd1aedec6428de08d8944b0344
