Update !487 and avoid implicit OpenMP parallelization in mesh::BoundarySetup

This merge request updates !487 (merged) where manual loops where exchanged with WALBERLA_FOR_ALL_CELLS macros. Since the macros are implicitly parallelized when using OpenMP, function setDomainCells() does not work as expected anymore. The content of vector domainCells will change when OpenMP parallelization is used such that its size can not be evaluated correctly in the later course of the algorithm. Thus OpenMP is turned off in the macro

This is the reason for the failure of the ComplexGeometry benchmark on the current master (see job 672374).