We now (once #1083 merged) fill magnetic ghosts but only use a static refinement whereas we could (should?) probably use a time intepolation.