Guiding Offline Reinforcement Learning Using Safety Expert