Seq-to-Seq & Transformer Deep Dive: From Encoder–Decoder to Modern LLMs