Come up with better hprod signature

[dpo]

Everywhere, hprod is defined as hprod(x, z, v) where z are Lagrange multipliers. This is cumbersome for unconstrained, bound-constrained or linearly-constrained problems. We should come up with a better signature and use it throughout. For example, hprod(v, x, z=None) or hprod(v, x, **kwargs) so unconstrained problems can simply use hprod(v, x). The vector to multiply with appears first, which is a bit counter-intuitive. In this case, for consistency, we should also use jprod(v, x) and jtprod(u, x) everywhere.