Enable symbolic Hessian-vector product

[rzyu45]

In some solvers, we need the derivative

$$\frac{\partial }{\partial y}\left(J(y)z\right)=H(y)\otimes z$$

where the Hessian tensor

$$H(y)=\left(\frac{\partial \nabla g_i(y)^\mathrm{T}}{y_j}\right)_{ij}.$$

We can avoid calculating the Hessian tensor by first multiplying out the Jac blocks so that we have

$$J(y)z= \begin{bmatrix} \sum_iJ_{1i}z_i\\\ \sum_iJ_{2i}z_i\\\ \cdots\\\ \sum_iJ_{ni}z_i \end{bmatrix}$$

Then we can obtain $\frac{\partial }{\partial y}\left(J(y)z\right)$ by deriving derivatives with respect to the right hand side.

With the new jac block class added in #76, the jac-vector product shall be divided into three categories:

1. if jac block is a matrix, then we have $J_{ij}$@ $z_i$;
2. if jac block is a diagonal matrix, then we have $\mathrm{diag}(J_{ij})$* $z_j$, where $\mathrm{diag}(J_{ij})$ is the vector of the diagonal entries of $J_{ij}$.
3. if jac block is a column vector, the we have $J_{ij}$* $z_j$.