CUDA error: unspecified launch failure when I changed the dispatch here

to AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND3
and ran this code on cuda:

import torch
a=torch.randn(2,3,dtype=torch.complex64).cuda()
b=a.to(torch.complex128)

RuntimeError: CUDA error: unspecified launch failure

cc @ezyang @anjali411 @dylanbespalko @ngimel