Check for __CUDA_ARCH_LIST__ for setting GT_CUDA_ARCH

[iomaganaris]

For some reason __CUDA_ARCH__ isn't set for the compilation of all intermediate files generated by nvcc. Meanwhile __CUDA_ARCH_LIST__ is.
This generates compilation errors for projects with GridTools (and more specifically ones that include storage/gpu.hpp) since code that needs to be generated for the GPU that is under a GT_CUDA_ARCH #ifdef isn't compiled.
Behavior was tested using CUDA 12.3 in various systems. Below is an example of the nvcc compilation command with the --verbose flag.

#$ gcc -std=c++20 -D__CUDA_ARCH__=600 -D__CUDA_ARCH_LIST__=600 -E -x c++  -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__  -I"/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/include" -I"/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/GridTools" -I"/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/build_gpu/_deps/gridtools-src/include" "-I/scratch/snx3000/ioannmag/spack/opt/spack/cray-cnl7-haswell/intel-2022.1.0/cuda-12.3.0-wagarpvzjr6mefjj42ym5343h3dketrg/bin/../targets/x86_64-linux/include"   -isystem "/scratch/snx3000/ioannmag/spack/opt/spack/cray-cnl7-haswell/intel-2022.1.0/boost-1.84.0-uw2w34r3fsrqlahxynev3rmv4oi55lq7/include" -isystem "/scratch/snx3000/ioannmag/spack/opt/spack/cray-cnl7-haswell/intel-2022.1.0/cuda-12.3.0-wagarpvzjr6mefjj42ym5343h3dketrg/targets/x86_64-linux/include"  -D "BOOST_PP_VARIADICS=1" -D__CUDACC_VER_MAJOR__=12 -D__CUDACC_VER_MINOR__=3 -D__CUDACC_VER_BUILD__=52 -D__CUDA_API_VER_MAJOR__=12 -D__CUDA_API_VER_MINOR__=3 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -include "cuda_runtime.h" -m64 "/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/src/wrapper.cpp" -o "wrapper.cpp1.ii"
#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=600 -E -x c++ -D__CUDACC__ -D__NVCC__  -I"/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/include" -I"/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/GridTools" -I"/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/build_gpu/_deps/gridtools-src/include" "-I/scratch/snx3000/ioannmag/spack/opt/spack/cray-cnl7-haswell/intel-2022.1.0/cuda-12.3.0-wagarpvzjr6mefjj42ym5343h3dketrg/bin/../targets/x86_64-linux/include"   -isystem "/scratch/snx3000/ioannmag/spack/opt/spack/cray-cnl7-haswell/intel-2022.1.0/boost-1.84.0-uw2w34r3fsrqlahxynev3rmv4oi55lq7/include" -isystem "/scratch/snx3000/ioannmag/spack/opt/spack/cray-cnl7-haswell/intel-2022.1.0/cuda-12.3.0-wagarpvzjr6mefjj42ym5343h3dketrg/targets/x86_64-linux/include"  -D "BOOST_PP_VARIADICS=1" -D__CUDACC_VER_MAJOR__=12 -D__CUDACC_VER_MINOR__=3 -D__CUDACC_VER_BUILD__=52 -D__CUDA_API_VER_MAJOR__=12 -D__CUDA_API_VER_MINOR__=3 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -include "cuda_runtime.h" -m64 "/scratch/snx3000/ioannmag/cycle21/icon_structured_benchmark/src/wrapper.cpp" -o "wrapper.cpp4.ii"
#$ -- Filter Dependencies -- > CMakeFiles/benchmark.dir/src/wrapper.cpp.o.d

For the use of gridtools::storage::gpu in the above example see https://github.com/GridTools/icon_structured_benchmark/blob/ioannmag/gpu/CMakeLists.txt, https://github.com/GridTools/icon_structured_benchmark/blob/ioannmag/gpu/include/nabla4_gridtools.hpp and https://github.com/GridTools/icon_structured_benchmark/blob/ioannmag/gpu/include/nabla4_structured_torus_gridtools_halo.hpp

[gridtoolsjenkins]

Hi there, this is jenkins continuous integration...
Do you want me to verify this patch?

[havogt]

replaced by #1778