Deep Learning From a Control Vantage Point: Feedforward, Residual, Transformers, Large Language Models, And Selective State-Space Models