Memory depth of finite state machine strategies for the iterated prisoner's dilemma