Transformer Decoder Block

Implement a NumPy Transformer decoder block with masked self-attention, cross-attention, and feed-forward layers.