Efficient Strategy Iteration for Mean Payoff in Markov Decision Processes