Distilling Event Sequence Knowledge From Large Language Models