Sample-efficient Reinforcement Learning Representation Learning with Curiosity Contrastive Forward Dynamics Model