KRAST: Knowledge-Augmented Robotic Action Recognition with Structured Text for Vision-Language Models