On Globally Optimal Punishments in the Repeated Cournot Game