Detecting Corpus-Level Knowledge Inconsistencies in Wikipedia with Large Language Models