Reinforcement Learning Makes LLMs Search-Savvy: Ant Group Researchers Introduce SEM to Optimize Tool Usage and Reasoning Efficiency