add frozen_numpy to :builtin_registry_cuda target (#67396)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/67396
frozen_numpy did not work on GPU since we didn't added register_frozennumpy to the :builtin_registry_cuda target.
This was not found earlier since the unit test we added to test_deploy.cpp is only run on CPU. On GPU, we run test_deploy_gpu.cpp which does not contains the added unit tests for numpy.
In this diff, I just duplidate the unit tests for numpy (and pyyaml) across test_deploy.cpp and test_deploy_gpu.cpp.
I think ideally we should consolidate there 2 files to a single one. So we can add unit test in a single place while run them in both hardward platforms.
Tracking task: T104399180
ghstack-source-id: 141750276
Test Plan: buck test mode/opt :test_deploy_gpu
Reviewed By: suo
Differential Revision: D31978156
fbshipit-source-id: 2f5cd55ca33107cc7d230b72f1353df81f0a3bda
