Learning to re-rank with Constrained Meta-Optimal Transport