Train full NumPy neural network on quality_binary

Combine the forward pass, backpropagation, and data batching into a complete training loop from scratch using only NumPy. Train the model on the scaled wine dataset for quality_binary. Plot the training loss curve and report the final test accuracy.