Migrate AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND1 to c10::complex (#39045)
Summary:
No special changes are needed for CPU kernels, some CUDA kernels are still doing `c10::complex -> thrust::complex` casting, this will be cleaned up later. But for now, it will be good to just keep it as is, and change the dispatch macro first.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39045
Differential Revision: D21741151
Pulled By: anjali411
fbshipit-source-id: 748f057f9f33338b8c9293aeaa228ad861172e71