Exposing Privacy Gaps: Membership Inference Attack on Preference Data for LLM Alignment