Train binary logistic regression from scratch using sigmoid probabilities, cross-entropy gradients, and gradient descent.
