Support custom constraints for `MovingHorizonEstimator`

[franckgaga]

As a first step, custom nonlinear inequality constrains should be implemented, and maybe custom linear inequality constraint as a second step.

In terms of API, the arguments of the custom nonlinear function $\mathbf{g_c}$ should have a physical meaning to the user, so it's better to left out the estimated noises:

1. The estimated states $\mathbf{\hat{x}}$ over $H_e$
2. The measured outputs $\mathbf{y^m}$ over $H_e$
3. The manipulated inputs $\mathbf{u}$ over $H_e$
4. The measured disturbances $\mathbf{d}$ over $H_e$
5. The parameter argument $\mathbf{p}$
6. The slack variable $\varepsilon$

The estimated process and sensor noises can be deduced from the above information, in the rare cases that the user would need them.