CarpetIOHDF5: do not recompute allactive data for output

instead use stored copy from dh::regrid. This speeds up output on many MPI ranks significantly as computing allactive is expensive.