Are Large Language Models Reliable Argument Quality Annotators?