Add dummy CUDA kernel for assert_async.msg (#101130)
This PR doesn't actually propogate the error messages down to internal kernel implementations. Will do in a follow up PR. This op is only really meant to be used in export context and export doesn't support CUDA right now so it is safe to just ignore the assert message. This PR is for quickly unblocking https://github.com/pytorch/pytorch/pull/100791 and https://github.com/pytorch/pytorch/issues/100918
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101130
Approved by: https://github.com/anijain2305, https://github.com/yanboliang
