Turn on TORCH_CHECK for NT wrap_buffer (#100596)
TORCH_INTERNAL_ASSERT_DEBUG_ONLY won't be enabled during non-debug builds, but for 1 dimension Tensors the check is cheap enough and not catching this can slow down development a lot.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100596
Approved by: https://github.com/drisspg
