VL-Mamba: Exploring State Space Models for Multimodal Learning