- allows for nested execution flow (i.e. solver loop inside time loop)
- timing tree integration
- parallel CUDA stream support

- WALBERLA_BUILD_WITH_OPENMP was never defined

- for example conda ships this old MPI version

- More flexibility for the user to implement custom checkpointing

MPICH2 apparently expects a non-const pointer for blocklenghts whereas
other MPIs expect const pointer -> removed const to work with all

- LB on GPU, Uniform Grid Benchmark app
- helper class to schedule tasks to multiple CUDA streams

- CUDA compiler has problems to compile external json library
- timing pool is included by BlockForest, and previously timing pool
  included JSON library

previously the ghost layer was aligned

Features:
   - uses generated pack infos for packing & unpacking directly on GPU
   - can directly send GPU buffers if cuda-enabled MPI is available,
     otherwise the packed buffers are transfered to CPU first
   - communication hiding with cuda streams: communication can be run
     asynchronously - especially useful when compute kernel is also
     split up into inner and outer part

- added RAII classes for CUDA streams and events
- equivalence test that checks if generated CPU and GPU (overlapped)
  versions are computing same result as normal waLBerla LBM kernel

- templated buffer classes allow the buffer systems to work with custom
  buffers, e.g. buffers of GPU memory