Text this: Spatiotemporal Deformation Prediction Model for Retaining Structures Integrating ConvGRU and Cross-Attention Mechanism