Multi-Agent Q-Learning Dynamics in Random Networks: Convergence due to Exploration and Sparsity