CUDA Codegen Error

Compiling the CUDA SDFG below yields a codegen error
(dace::GlobalToGlobal cannot be found)

[global_to_global_bug.sdfg.zip](https://github.com/spcl/dace/files/5036189/global_to_global_bug.sdfg.zip)

Instead of generating `GlobalToGlobal` in `dace/codegen/targets/cuda.py`, we should raise a `NotImplementedError` that mentions that GPU global memory to global memory copies need to be more explicitly specified in the code.

To reproduce this (and we should add a test), we can add a `GPU_Global` access node outside and a `GPU_Global` access node inside a `GPU_Device` map.