Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning