Proto Successor Measure: Representing the Space of All Possible Solutions of Reinforcement Learning