Improve the error message for VectorNonlinearFunction

[amontoison]

close #204
We need a JuMP model with a SNF as objective and a VNF for the constraints for the tests.

[amontoison]

@blegat
It seems that your code to add a VectorNonlinearFunction in the model is not working here.
I checked the documentation and the function MathOptInterface.Nonlinear.add_constraint is not defined for a VNF.

cc @odow

[blegat]

We could also use bridges in MathOptNLPModel(model). Making sure to only select bridges that do not add variables:
https://github.com/JuliaPolyhedra/Polyhedra.jl/blob/master/src/lphrep.jl#L40-L91

[amontoison]

@blegat For the bridges, do you have one that reformulate the complementarity constraints x \ perp y into x >=0, y>=0, xy == 0?

[odow]

Nope. It's on my TODO list for a separate package. Also note that your definition of complementarity does not match MOI's. The restrictions on x depend on the domain of y.

[amontoison]

We use this reformulation in our case for SCOPF:
JuliaSmoothOptimizers/AmplNLReader.jl#144

I should add a proper support of these constraints in the API NLPModels.jl such that the hybrid ExaModels.jl can exploit what you did in MOI.