LongVQUBench: Benchmarking Long-Term Video Quality Understanding of Vision-Language Models