build on nvidia machines with enzyme

In order to build enzyme on nvidia-based machines (e.g. matrix), we need to set the CMAKE_CUDA_COMPILER to LLVM rather than NVCC. This has some minor inconveniences such as needing to remove the `-Xlinker`.

The issue I got stuck on was despite setting the cuda compiler to LLVM, it still used nvlink during the final link step, causing this issue:

> nvlink error : Undefined reference to 'enzyme_interleave' in 'examples/explicit/CMakeFiles/explicit_dfem_gpu.dir/dfem_gpu.cpp.o'
make[3]: ***

I believe we need to set the cuda linker somehow to something other than nvlink.