Revisiting the Minimalist Approach to Offline Reinforcement Learning