Enhancing LLM-based Hatred and Toxicity Detection with Meta-Toxic Knowledge Graph