RDRec: Rationale Distillation for LLM-based Recommendation