A note on overrelaxation in the Sinkhorn algorithm