Probing Ranking LLMs: Mechanistic Interpretability in Information Retrieval