Build support for Cray toolchains with AMD GPUs

[ohearnk]

This PR fixes build support for Cray toolchains with AMD GPUs. Specifically, this PR is a work-in-progress for running QUICK on the SDSC Cosmos system (AMD 300A).

Currently, the single GPU (HIP) code builds and runs successfully with the following commands:

# allocate job on and log in to main cluster node using Slurm
module swap PrgEnv-cray/8.5.0 PrgEnv-gnu-amd/8.5.0
module load craype-accel-amd-gfx942
module unload cray-mpich/8.1.30 cray-libsci/24.07.0   # TODO: fix MPI+HIP builds with Cray MPI and math libs
cmake .. -DCMAKE_BUILD_TYPE=RELEASE -DOPTIMIZE=TRUE -DCOMPILER=MANUAL -DCMAKE_C_COMPILER=amdclang -DCMAKE_CXX_COMPILER=amdclang++ -DCMAKE_Fortran_COMPILER=gfortran-13 -DCMAKE_C_FLAGS="-O2" -DCMAKE_CXX_FLAGS="-O2" -DCMAKE_Fortran_FLAGS="-O2 -ftree-vectorize -funroll-loops -ffast-math" -DMPI=FALSE -DHIP=TRUE -DENABLEF=FALSE -DQUICK_USER_ARCH=gfx942 -DFORCE_INTERNAL_LIBS="lapack;blas" -DCMAKE_INSTALL_PREFIX=${PWD}/../install_master_PrgEnv-gnu-amd8.5.0_gcc-native13.2.0_rocm6.3.0 &> cmake_configure.log
cmake --build . --verbose &> cmake_build.log
cmake --install .

Merge this only once simplified Cray builds are fixed (via the -DCOMPILER=CRAY CMake variable) and other build issues are resolved (MPI+HIP builds with Cray MPI, Cray math libs, etc.). Note also that support for Cray builds with upstream Ambertools need to be synchronized (hence, patches may need to be upstreamed there for HIP support).