MSE Loss Forward and Backward

Implement mean squared error over all elements and its gradient with respect to the predictions:

• mse(pred, target) = $\frac{1}{N}\sum (p - t)^2$ where $N$ is the total number of elements
• mse_grad(pred, target) = $\frac{2}{N}(p - t)$, same shape as pred