EVE: Efficient Vision-Language Pre-training with Masked Prediction and Modality-Aware MoE