SimSR: Simple Distance-based State Representation for Deep Reinforcement Learning