Simplify work with walberla kernels

[RudolfWeeber]

When adding new kernels, a lot of manual work is currently needed

• write the actual kernel generation in maintainer/walberla_kernels/generate_*.py
• generate, format and copy kernels to src/walberla_bridge/src/lattice_boltzmann/generated_kernels

the following have to be done 4x per kernel, single/double precision, cpu/gpu

• Add kernel cpp files to the CMake file there
• include headers of the new kernels in kernel_traits.hpp
• add the kernel classes in the KernelTraits classes for the respective architectures
• lastly, kernels are currently imported into the actual LBWalberlaImpl class template via a using statement.

There is probably some potential for streamlingin:

• the various kernel generation steps described in the readme in maintainer/walberla_kernels can probably mostly be put into a shell script
• at the time of kernel generation, all information needed for the kernel_traits.hpp is in princicle known. We could generate that file from a Jinja template
• Having one using statement per Kernel in LBWalberlaImpl could also be avoided, this would, however result in extra typenames throughout the code. Might still be worth it.

using Kernels = detail::KernelTraits<FloatType, Architecture>;

// ...
auto m_initial_pdf_setter = std::make_shared<typename Kernels::InitialPDFSetter>;