HOI-R1: Exploring the Potential of Multimodal Large Language Models for Human-Object Interaction Detection