Through the Lens of Core Competency: Survey on Evaluation of Large Language Models