Visual Attention Model for Cross-Sectional Stock Return Prediction and End-to-End Multimodal Market Representation Learning