Set number of threads to be 1 for ARM (#97482) (#98267)
Summary:
In highly multi-threaded environment, using # of threads to be matching hardware_concurrency leads to high contention. x86 path actually ends up using different path (MKL path), which results in using 1 thread for x86 as well.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98267
Approved by: https://github.com/malfet