Monte Carlo Sampling for Regret Minimization in Extensive Games