[BUG] Failing tests on GB300

**Describe the bug**
Many cudf tests fail on GB300 due to CUDA address misalignment issues and other errors.

**Steps/Code to reproduce bug**
Run the tests on GB300

**Expected behavior**
All tests should pass

**Environment details**
GB300 CI runner

**Additional context**
Failing tests can be seen here:
https://github.com/rapidsai/cudf/actions/runs/30636553697?pr=23246

We should start with `conda-cpp-tests`, and this may resolve the other issues:
https://github.com/rapidsai/cudf/actions/runs/30636553697/job/91185474271?pr=23246

Start by examining the failing conda-cpp-tests job in the linked CI run and reproduce the failures on the GB300 CI runner. Investigate the reported CUDA address misalignment and other errors, then verify that conda-cpp-tests and the remaining cuDF tests pass on GB300.

Written by the indexing model from the issue text.