Back out "[CUDA Pinned Memory] Alternative implementation of pinned memory allocator focusing on multi-threaded scalability" (#69191)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/69191
Reviewed By: xing-liu, yuchenhao
Differential Revision: D32748466
fbshipit-source-id: 6abd3265e8a20270305da3f8be25114ad4d67fc2