ISO-Bench: Benchmarking Multimodal Causal Reasoning in Visual-Language Models through Procedural Plans