D2RL: Deep Dense Architectures in Reinforcement Learning