RePo: Resilient Model-Based Reinforcement Learning by Regularizing Posterior Predictability | Read Paper on Bytez