Size and depth of monotone neural networks: interpolation and approximation