Fighting FIRe with FIRE: Assessing the Validity of Text-to-Video Retrieval Benchmarks