[BUG] unbounded problem solved at presolve wrongly reported infeasible

[mtanneau]

Describe the bug

A problem found by PSLP to be "infeasible or unbounded" is wrongly reported to be infeasible.

I encountered this issue on v26.02 via the Julia and python APIs.

Steps/Code to reproduce bug

In python:

from cuopt.linear_programming.problem import Problem, CONTINUOUS, MINIMIZE
from cuopt.linear_programming.solver_settings import SolverSettings

problem = Problem(name="unbounded")
x = problem.addVariable(lb=0.0, vtype=CONTINUOUS, name="x")
y = problem.addVariable(lb=0.0, vtype=CONTINUOUS, name="y")

problem.addConstraint(-1*x + 2*y <= 0, name="c1")

problem.setObjective(-1*x - 1*y, sense=MINIMIZE)

settings = SolverSettings()

problem.solve(settings)

problem.Status.name  # 'PrimalInfeasible'

This problem corresponds to the MPS file below:

NAME
ROWS
 N  OBJ
 L  c1
COLUMNS
    x        OBJ       -1
    x        c1        -1
    y        c1        2
    y        OBJ       -1
RHS
    rhs       c1        0
RANGES
BOUNDS
 LO bounds    x        0
 LO bounds    y        0
ENDATA

Here is the log I obtain from cuOpt:

cuOpt version: 26.2.0, git hash: f73da24d, host arch: aarch64, device archs: 75-real,80-real,86-real,90a-real,100f-real,120a-real,120
CPU: Unknown, threads (physical/logical): 1/20, RAM: 107.53 GiB
CUDA 13.1, device: NVIDIA GB10 (ID 0), VRAM: 119.70 GiB
CUDA device UUID: 43bce240-d7d6-65e5-ca7a-e351bab5de7a

Solving a problem with 2 constraints, 4 variables (0 integers), and 4 nonzeros
Problem scaling:
Objective coefficents range:          [1e+00, 1e+00]
Constraint matrix coefficients range: [1e+00, 2e+00]
Constraint rhs / bounds range:        [0e+00, 0e+00]
Variable bounds range:                [0e+00, 0e+00]

Using PSLP presolver
PSLP declares problem as infeasible or unbounded.
PSLP Presolved problem: 0 constraints, 0 variables, 0 non-zeros

Note that PSLP correctly reports the problem as "infeasible or unbounded", but cuOpt returns only the "infeasible" part.

Expected behavior

The problem is unbounded: (x, y) = (0, 0) is feasible, and the ray (x, y) = (1, 0) is feasible and has negative objective value. cuOpt should return that the problem is unbounded / dual infeasible.

Environment details (please complete the following information):

• Environment location: DGX Spark with GB10
• Method of cuOpt install: in a python virtual environment using uv.

Additional context

cuOpt does not have a "dual infeasible" termination status, which would be appropriate here.