SealQA: Raising the Bar for Reasoning in Search-Augmented Language Models