This was not working, as kernelNeedsFluidNodeIndicesToRun is used before
creating the kernel. I still don't like the current solution.

This reverts commit 273156ba.

changed the selection of Kernels - now we use the variables from KernelType.h in the parameter class an some apps

- renamed the mpi module to parallel
- removed the communicationRoutine interface from gpu. gpu now also uses the communicator interface from parallel
- added missing functionality in parallel/Communicator
- moved ws2_32 library dependency on windows from VirtualFluids_GPU to parallel

Extract Scaling and interpolation from CF Interpolation to generalize this behavior for CPU and GPU.

[GPU] Fix typo, in Probe.cu we used "part" instead of "Part", which results in failing regression tests.