This does not yet use the pystencils' cache folder or disk caching of the compilation.
This can be used to embed compiled bytecode into waLBerla executables as I do with my Vulkan wrapper. Not sure if this is a good way to go but at least we can experiment with it.
A good way to proceed with this MR is also a comparison between hip/sicl/ocl/vulkan in order to identify a suitable backend for pystencils.
