[BUG]: free-threaded python is flaky, blocking CI

[jpascucci-nv]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Something else

Component

General cuda-python

Describe the bug

Free-threaded cuda-python CI from main is failing sometimes on 3.14t and sometimes on 3.15t.

Here's an example of a failure on 3.14t.
Here's an example on 3.15t.

How to Reproduce

It appears to happen with any change, for instance, my PR really only affects CI.

Expected behavior

I expect main to run CDCI successfully.

Operating System

No response

nvidia-smi output

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.84                 Driver Version: 595.84         CUDA Version: 13.2     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA L4                      On  |   00000000:0A:00.0 Off |                    0 |
| N/A   40C    P8             13W /   72W |       0MiB /  23034MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

[mdboom]

This is coming from an agent-written test test_allocation_handle_forking_pickler_roundtrip, added to increase coverage, that was merged yesterday. I'm not certain it's even testing something that is explicitly supported, but maybe @Andy-Jost has an insight.

Given that, I think it would be fine to skip the test while we investigate if there is a fix to some root cause that needs to be applied.

[jpascucci-nv]

duplicate of #2784