CARPAS: Towards Content-Aware Refinement of Provided Aspects for Summarization in Large Language Models