Truncating Trajectories in Monte Carlo Reinforcement Learning