Stochastic mathematical programs with probabilistic complementarity constraints: SAA and distributionally robust approaches