Near-Optimal Learning of Extensive-Form Games with Imperfect Information