Dual Interior-Point Optimization Learning