Classifying Unreliable Narrators with Large Language Models