In the latter case, the positions themselves are not equivalent, presumably too large a tolerance was used in the symmetrization routine (and different tolerance for volume and Nsites determination).
However, the former case, just as the one in the OP (mp-19009), is AFM, so a correction to Nsites is non-trivial and due to the true (magnetic) unit cell being twice the unit cell of the atomic structure… Sorry if I’m saying something obvious, just wanted to draw your attention to this potential complication in fixing the problem.