S2R-ViT for Multi-Agent Cooperative Perception: Bridging the Gap from Simulation to Reality