Computes gradient with respect to the natural parameters.

The function takes the Riemannian gradient as an input. This is for three reasons: 1) You probably want to use the Riemannian gradient anyway so this helps avoiding accidental use of this function. 2) The gradient is computed by using the Riemannian gradient and chain rules. 3) Probably you need both Riemannian and normal gradients anyway so you can provide it to this function to avoid re-computing it.