Disagreeing Rationales: Rethinking Classification and Explainability Evaluation in Hate Speech Detection — ThinkLLM