Stochastic Latent Actor-Critic: Deep Reinforcement Learning with a Latent Variable Model