Watch and Match: Supercharging Imitation with Regularized Optimal Transport