Create policy for having CMAKE_CUDA_ARCHITECTURES default to all GPUs on the system
When we compile typical code (C, C++, etc.) - it is assumed, and CMake assumes, that we are targeting the CPU of the build system. This should also be the case for CUDA code: it is reasonable to assume that the targeted GPU(s) are those present on the system.
Now, I'm hesitant to suggest that this be detected by default, since detection has a cost, but at least there should be an optional policy for such auto-detection, which would result in the equivalent of
set(CMAKE_CUDA_ARCHITECTURES AUTO)
(This is not yet implemented, see issue #22375 (closed))