Convergence of Optimistic and Incremental Q-Learning