use CUDA 13 for third-party integration tests

[jameslamb]

Description

Contributes to rapidsai/build-planning#208

Reverts #19903 and #19917

Now that there are xgboost conda packages with CUDA 13 support (rapidsai/xgboost-feedstock#100), switches the third-party-integration-tests-cudf-pandas to use CUDA 13.

This also switches to the more generic CI Image tags (rapidsai/26.02-latest), so that configuration doesn't need to be updated as RAPIDS adds support for new minor versions of CUDA.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[jameslamb]

/ok to test

[jameslamb]

I think this is working... the third-party-integration-tests-cudf-pandas job is passing: https://github.com/rapidsai/cudf/actions/runs/19827202930/job/56806558906?pr=20748

Other failures look like network or Anaconda infrastructure issues, which would hopefully be resolved with a re-run:

critical libmamba Multiple errors occurred:
Download error (28) Timeout was reached [https://conda.anaconda.org/conda-forge/noarch/repodata.json.zst]
Operation too slow. Less than 30 bytes/sec transferred the last 60 seconds
Subdir conda-forge/noarch not loaded!
If you run into this error repeatedly, your package cache may be corrupted.
Please try running mamba clean -a to remove this cache before retrying the operation.
If you still are having issues, please report the error on mamba-org/mamba's issue tracker:
https://github.com/mamba-org/mamba/issues/new?assignees=&labels=&projects=&template=bug.yml
[rapids-conda-retry] conda returned exit code: 1
[rapids-conda-retry] Exiting, reached max retries...

(conda-python-other-tests link)

[vyasr]

I've had #19948 open for a while on this but was stalled waiting for Pytorch. In the interim (last week) we merged #20729, which means that we now accept installing CPU pytorch since we then just skip those tests. @Matt711 you pretty much said you're fine with this outcome, right? Like @bdice said, there's some risk of us just not noticing happening, but I agree that it's a net benefit overall.

[bdice]

@jameslamb This should be good to merge once the temporary channel changes are removed. Thanks for testing that!

[jameslamb]

Ah thanks @vyasr , sorry I missed that already-open PR.

I've removed the other testing bits from this, it should be ready now.

[jameslamb]

/merge