What is the Transformer architecture in Large Language Models, and why is it important for processing language data? How do self-attention, embeddings, positional information, and multiple neural network layers help Transformers understand relationships between tokens?