SVAgent: Storyline-Guided Long Video Understanding via Cross-Modal Multi-Agent Collaboration | Read Paper on Bytez