Benchmarking Vision-Language Models for Road Scene Question Answering under Day and Night Conditions | Synapse