HateXplain: A Benchmark Dataset for Explainable Hate Speech Detection