Optimal paths in multi-stage stochastic decision networks