How to solve large scale deterministic games with mean payoff by policy iteration