PuLP test_infeasible_2: solver returns Optimal for an infeasible LP

Summary

pulp/tests/test_pulp.py::CUOPTTest::test_infeasible_2 fails in the nightly CI because cuOpt reports status=1 (Optimal) for an LP that is genuinely infeasible. Expected -1 (Infeasible).

Observed in the nightly run on 2026-04-16 (wheel-python-cuda13.1.1-py3.14-aarch64, cuOpt 26.6.0, git hash f72f57da).

This is distinct from #1114 — it is not a status-mapping issue. The solver is producing a wrong terminal status on the underlying problem.

Reproducer (from the failing test)

from pulp import LpProblem, constants as const
from pulp.apis.cuopt_api import CUOPT

prob = LpProblem("test_infeasible_2", const.LpMinimize)
x = prob.add_variable("x", 0, 4)
y = prob.add_variable("y", -1, 1)
z = prob.add_variable("z", 0, 10)
prob += x + y <= 5.2, "c1"
prob += x + z >= 10.3, "c2"
prob += -y + z == 17.5, "c3"
status = prob.solve(CUOPT())
assert status == const.LpStatusInfeasible  # currently reports Optimal

Why the problem is infeasible

Constraint c3: -y + z = 17.5 with y ∈ [-1, 1], z ∈ [0, 10] gives max(-y + z) = 1 + 10 = 11 < 17.5. The constraint is not satisfiable, so the LP is infeasible.

Failure output

E   pulp.constants.PulpError: Tests failed for solver <pulp.apis.cuopt_api.CUOPT ...>:
E   status == 1 not in [-1]
E   status == Optimal not in ['Infeasible']

Context

• Several other unbounded/infeasible cases in the same run fail with status-11 mapping issues (tracked in #1114). This one is different: cuOpt is actually returning Optimal, not the new UnboundedOrInfeasible status.
• Worth confirming whether this reproduces on the LP alone (via cuopt_cli on an MPS export) to localize whether it's a solver regression or a PuLP-adapter model-translation bug.