ShotBench: Expert-Level Cinematic Understanding in Vision-Language Models | Read Paper on Bytez