Dual infeasibility returned for feasible problem with box constraints

[migarstka]

I stumbled upon this problem from the MathOptInterface testset

 maximize x + y
 s.t.  2 <= x + y <= 12
          x,  y >= 0

When solved with a box constraint, COSMO returns dual infeasible, which is wrong.

using COSMO, SparseArrays, LinearAlgebra

q = [-1; -1.]
P = spzeros(2, 2)

# x,  y >= 0
nn_constraint = COSMO.Constraint(Matrix(1.0I, 2, 2), zeros(2), COSMO.Nonnegatives)

# 2 <= x + y <= 12
box_constraint = COSMO.Constraint([1. 1], zeros(1), COSMO.Box([2.], [12.]))

cs = [nn_constraint; box_constraint]
settings = COSMO.Settings(verbose=true, eps_abs = 1e-5, eps_rel = 1e-5, check_termination = 25, check_infeasibility = 25)

model = COSMO.Model()
assemble!(model, P, q, cs, settings = settings)
res = COSMO.optimize!(model);

However, when implemented with one-sided constraints COSMO solves the problem.

nn_instead_of_box = COSMO.Constraint([1. 1.; -1. -1], [2; 12], COSMO.Nonnegatives)
cs = [nn_constraint; nn_instead_of_box]

model = COSMO.Model()
assemble!(model, P, q, cs, settings = settings)
res = COSMO.optimize!(model);

I think this is probably an issue with in_pol_recc() for COSMO.Box.