Similar Items: Stack Transformer-Based Spatial-Temporal Attention Model for Dynamic Sign Language and Fingerspelling Recognition