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.

Locate the notebook or entry point containing the existing forward pass, backpropagation, batching, and scaled wine dataset, then read and run those pieces first. Assemble them into a complete NumPy training loop for quality_binary, plot the training loss, and report final test accuracy; completion is demonstrated by the curve and accuracy result.

Written by the indexing model from the issue text.