An Investigation of Prompt Variations for Zero-shot LLM-based Rankers