Convergence of Actor-Critic Learning for Mean Field Games and Mean Field Control in Continuous Spaces