Through a PRISM, Clearly: Step-Sequence Reasoning for Mistake Recognition in Procedural Videos