Investigating Annotator Bias in Large Language Models for Hate Speech Detection