Support `pvv` terms in objective

[klamike]

Currently

using JuMP, Dualization
m = Model()
@variable m x
@variable m p ∈ Parameter(2.0)
@objective m Min 3 * x * x * p
print(dualize(m, dual_names=DualNames()))

errors with Objective functions of type MathOptInterface.ScalarNonlinearFunction are not implemented,
but it can (should?) be

Max -3 quadslack_x² * param_p
Subject to
 x : -3 quadslack_x * param_p = 0

Note the parametric QP formulation in the docs does not allow for this (i.e. it is consistent with the current behavior).

[odow]

Because of the way JuMP represents parameters and quadratic expressions, this issue unlikely to be implemented.

[odow]

Closing as won't-fix. This is beyond the scope of Dualization.jl.

[joaquimg]

We could support some subset of ScalarNonlinearFunction in the objective. We should do it first in ParametricOptInterface.jl jump-dev/ParametricOptInterface.jl#144

Once it is fully settled there, we can consider something here. Doing it here would be considerably harder.