MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents - Explained Simply | ArXiv Explained