Text this: Watermark and Trademark Prompts Boost Video Action Recognition in Visual-Language Models