LEMMA: Towards LVLM-Enhanced Multimodal Misinformation Detection with External Knowledge Augmentation