add test sharding to CUDA on linux (#45972)
Summary:
splits up all the cuda linux tests into 2 shards to decrease total test runtime
Pull Request resolved: https://github.com/pytorch/pytorch/pull/45972
Reviewed By: malfet
Differential Revision: D24163521
Pulled By: janeyx99
fbshipit-source-id: da6e88eb4305192fb287c4458c31199bf62354c0