defining host and device when backend != cuda
Since https://github.com/NVIDIA/cccl/pull/1199 these macros are undefined