This PR introduces custom nonlinear inequality constraints for LinMPC and NonLinMPC objects. The constructor of both objects now accepts 4 new keyword arguments:

• Wy=nothing : custom linear constraint matrix for output.
• Wu=nothing : custom linear constraint matrix for manipulated input.
• Wd=nothing : custom linear constraint matrix for meas. disturbance.
• Wr=nothing : custom linear constraint matrix for output setpoint.

and the setconstraint! method now accept these keyword arguments:

• wmin=fill(-Inf,nw) / wmax=fill(+Inf,nw) : custom linear constraint bound $\mathbf{w_{min/max}}$
• c_wmin=fill(1.0,nw) / c_wmax=fill(1.0,nw) : wmin / wmax softness weight $\mathbf{c_{w_{min/max}}}$
• the keyword arguments above but with a first capital letter, e.g. Wmax or C_wmin: for time-varying constraints

The custom nonlinear inequality constraints are defined as a linear combination of the output prediction $\mathbf{\hat{y}}$, manipulated input $\mathbf{u}$, measured disturbance prediction $\mathbf{\hat{d}}$ and setpoint prediction $\mathbf{\hat{r}_y}$, as described in the extended help section of setconstraint! :

I chose to incorporate the output prediction $\mathbf{\hat{y}}$ instead of state prediction $\mathbf{\hat{x}}$ in them mainly because its associated prediction matrices are already constructed and available inside LinMPC and NonLinMPC objects. This is non-limiting in practice since you can always output the states that are needed by modifying the state-space $\mathbf{C}$ and $\mathbf{D_d}$ matrices, and adjusting the i_ym argument of the state estimator if these states are not measured.

Note that C code generation is not supported for now but I will add the support in the following PR.