Learning to Re-rank with Constrained Meta-Optimal Transport