Implement forward pass and activation from scratch

Write pure NumPy functions for a dense layer's forward pass (Z = WX + b), and implement ReLU (for the hidden layer) and Sigmoid (for the output layer) activation functions. Test them with dummy weights and a small batch of wine data to ensure matrix shapes align perfectly.

Start in the notebook cells or entry points that define the dense forward pass and activation functions. Use the dummy weights and small wine-data batch described in the issue to check matrix shapes, then verify ReLU for the hidden layer and Sigmoid for the output layer produce the expected results.

Written by the indexing model from the issue text.