Fix compile and log

[akifcorduk]

This PR fixes a log level and a compilation error that happens on non-CI envs.

[hlinsen]

Thanks Akif. Do you think we could also disable postsolve logs or change the log level https://github.com/NVIDIA/cuopt/blob/branch-25.10/cpp/src/mip/presolve/third_party_presolve.cpp#L311?

[akifcorduk]

@hlinsen IMO, we can keep it. The user would know why we are returning a slightly infeasible solution. But I think we can get @chris-maes opinion too.

[rgsl888prabhu]

@akifcorduk not sure why it is failing or even where it is failing, but this test run has been failing several times, I think it has been run atleast 4 times. Please take a look in earlier run if you can find any clue https://github.com/NVIDIA/cuopt/actions/runs/18288323659/job/52087939557?pr=450

On the top you can see attempts, where you can switch to previous runs.

[nguidotti]

Some test runs returns a SIGSEGV, but their location depends on each attempt...

Latest:

 [ RUN      ] level0_ges/double_test_vrp.GES_VRP/0
double free or corruption (out)
timeout: the monitored command dumped core
ci/test_cpp.sh: line 61:  1634 Aborted                 timeout 20m "${gt}" --gtest_output=xml:"{RAPIDS_TESTS_DIR}"

3rd attempt:

 [ RUN      ] pdlp_class.test_max_with_offset
cuOpt version: 25.10.0, git hash: 2fae644, host arch: x86_64, device archs: 70-real,75-real,80-real,86-real,90a-real,100f-real,120a-real,120
CPU: AMD EPYC 9554 64-Core Processor, threads (physical/logical): 12/12, RAM: 42.45 GiB
CUDA 12.2, device: NVIDIA L4 (ID 0), VRAM: 21.96 GiB
CUDA device UUID: ffffff984a044b-0337-ffffff80ffffffc5

Third-party presolve is disabled, skipping
Solving a problem with 1 constraints 3 variables (0 integers) and 3 nonzeros
Objective offset 4.000000 scaling_factor -1.000000
   Iter    Primal Obj.      Dual Obj.    Gap        Primal Res.  Dual Res.   Time
      0 -4.00000000e+00 -4.00000000e+00  0.00e+00   0.00e+00     3.74e+00   0.003s
    210 -7.25214461e-05 -7.23309242e-05  1.91e-07   0.00e+00     1.23e-04   0.012s
LP Solver status:                Optimal
Primal objective:                -7.25214461e-05
Dual objective:                  -7.23309242e-05
Duality gap (abs/rel):           +1.91e-07 / +1.90e-07
Primal infeasibility (abs/rel):  +0.00e+00 / +0.00e+00
Dual infeasibility (abs/rel):    +1.23e-04 / +2.60e-05
PDLP finished
Status: Optimal   Objective: -7.25214461e-05  Iterations: 210  Time: 0.012s, Total time 0.013s
[       OK ] pdlp_class.test_max_with_offset (17 ms)
[ RUN      ] pdlp_class.test_lp_no_constraints
cuOpt version: 25.10.0, git hash: 2fae644, host arch: x86_64, device archs: 70-real,75-real,80-real,86-real,90a-real,100f-real,120a-real,120
CPU: AMD EPYC 9554 64-Core Processor, threads (physical/logical): 12/12, RAM: 42.45 GiB
CUDA 12.2, device: NVIDIA L4 (ID 0), VRAM: 21.96 GiB
CUDA device UUID: ffffff984a044b-0337-ffffff80ffffffc5

Third-party presolve is disabled, skipping
Solving a problem with 0 constraints 1 variables (0 integers) and 0 nonzeros
Objective offset -0.000000 scaling_factor -1.000000
Running concurrent

No constraints in the problem: PDLP can't be run, use Dual Simplex instead.
PDLP finished
Status: A numerical error was encountered.   Objective: nan  Iterations: -1  Time: 0.000s, Total time 0.001s
Handling removed variables 1
Dual simplex finished in 0.00 seconds, total time 0.00
terminate called after throwing an instance of 'thrust::system::system_error'
  what():  parallel_for failed: cudaErrorInvalidConfiguration: invalid configuration argument
timeout: the monitored command dumped core
ci/test_cpp.sh: line 61:  1595 Aborted                 timeout 20m "${gt}" --gtest_output=xml:"{RAPIDS_TESTS_DIR}"

[akifcorduk]

/merge