A New Perspective on Transformers in Online Reinforcement Learning for Continuous Control