Experimental: picurv run takes the MPI decomposition from optional --ni/--nj/--nk, from cluster.yml execution.decomposition, or chooses it from the grid, its multigrid levels and periodicity when only -n is given; the choice is printed before launch and in --dry-run, the postprocessor gets its own, and case.yml grid.da_processors_x/y/z is deprecated but still read.