LLM in the Loop: Creating the ParaDeHate Dataset for Hate Speech Detoxification