Running with -zero and axes 0,1,2 together in the one call on a 4D image segfaults.

On a related note: Never been a fan of the code branching in mrconvert. Could the -zero option be handled in mrcalc instead?