ChromouVQA: Benchmarking Vision-Language Models under Chromatic Camouflaged Images