Exploring the Potential of Large Multimodal Models as Effective Alternatives for Pronunciation Assessment