Reductions on HIP

Reductions are so far only implemented for the CUDA GPU target (!438 (merged)!). They shall also be made available for the HIP platform.

Tasks

• Implement resolution of per-thread and warp-level reductions in HipPlatform
• Implement handling of warp sizes for AMD GPUs

Prerequesites

It would be nice to have a CI task for testing pystencils on HIP + AMD Hardware, e.g. using the testcluster, before integrating this feature.