CoSpace: Benchmarking Continuous Space Perception Ability for Vision-Language Models | Read Paper on Bytez