LR Range Test and One-Cycle Policy

Implement the one-cycle learning rate policy for full-batch MSE gradient descent, using zero-initialized weights.