Variable bounds not included in list of constraint indices

[LukasBarner]

Currently, model.ctr_lower does not include variable bound references:

using JuMP, BilevelJuMP, Ipopt
model = BilevelModel(Ipopt.Optimizer, mode = BilevelJuMP.ProductMode())
@variable(Upper(model), x)
@variable(Lower(model), y>=0)
@objective(Upper(model), Min, -x + 4y)
@constraint(Upper(model), y + 2x <= 9)
@objective(Lower(model), Min, -x - y )
@constraint(Lower(model), x + y <= 8)
@constraint(Lower(model),  x <= 4)
optimize!(model)
model.ctr_lower
# Dict{Int64, ConstraintRef} with 2 entries:
#   4 => x ≤ 4.0
#   3 => x + y ≤ 8.0
model.lower_dual_to_sblm
# MathOptInterface.Utilities.IndexMap with 7 entries:
#   VariableIndex(3)                                                    => VariableIndex(3)
#   VariableIndex(1)                                                    => VariableIndex(4)
#   VariableIndex(2)                                                    => VariableIndex(5)
#   ConstraintIndex{VariableIndex, LessThan{Float64}}(2)                => ConstraintIndex{VariableIndex, LessThan{Float64}}(5)
#   ConstraintIndex{VariableIndex, LessThan{Float64}}(1)                => ConstraintIndex{VariableIndex, LessThan{Float64}}(4)
#   ConstraintIndex{VariableIndex, GreaterThan{Float64}}(3)             => ConstraintIndex{VariableIndex, GreaterThan{Float64}}(3)
#   ConstraintIndex{ScalarAffineFunction{Float64}, EqualTo{Float64}}(1) => ConstraintIndex{ScalarAffineFunction{Float64}, EqualTo{Float64}}(1)
model.lower_primal_dual_map.primal_con_dual_var
# Dict{MathOptInterface.ConstraintIndex, Vector{MathOptInterface.VariableIndex}} with 3 entries:
#   ConstraintIndex{ScalarAffineFunction{Float64}, LessThan{Float64}}(2) => [VariableIndex(2)]
#   ConstraintIndex{VariableIndex, GreaterThan{Float64}}(2)              => [VariableIndex(3)]
#   ConstraintIndex{ScalarAffineFunction{Float64}, LessThan{Float64}}(1) => [VariableIndex(1)]

This leads to unexpected behavior when passing starting values to duals of variable bounds, they are not passed in this code sequence:

BilevelJuMP.jl/src/jump.jl

Lines 515 to 524 in 2d25399

	for (idx, info) in model.ctr_info
	if haskey(model.ctr_lower, idx)
	ctr = model.ctr_lower[idx]
	# this fails for vector-constrained variables due dualization 0.3.5
	# because of constrained variables that change the dual
	pre_duals = lower_primal_dual_map.primal_con_dual_var[JuMP.index(ctr)] # vector
	duals = map(x->lower_dual_to_sblm[x], pre_duals)
	pass_dual_info(single_blm, duals, info)
	end
	end

Fixing this would be possible by:

a) adding lower level variable bounds to model.ctr_lower after the single_blm is built. The existing code could do the rest of the work as model.lower_dual_to_sblm and model.lower_primal_dual_map.primal_con_dual_var already include the bounds.
b) Passing start values for all constrained variables after the single_blm is built.

Approach b) works fine (and I guess hints could be added the same way), for a) I am not sure if this might mess other things up that I didn't notice...
If a) does not cause chaos on other ends, I think it should be preferred?
I could take care of this in a PR.

[LukasBarner]

May be obsolete if #163