Theoretical and Algorithmic Foundations of In-Context Learning and Chain of Thought Using Properly Trained Transformer Models