[BUG] hash_aggregate_test.py::test_hash_multiple_grpby_pivot and row_conversion_test.py::test_row_conversions_fixed_width_wide fails on GB100 and cuda12.8

[yinqingh]

Describe the bug
Observed following similar failures in the test on GB100 with cuda12.8

• hash_aggregate_test.py::test_hash_multiple_grpby_pivot

Exception in task 0.0 in stage 35.0 (TID 90) ai.rapids.cudf.CudfException: CUDA error at: build-spark-rapids-jni-cuda12/target/libcudf/cmake-build/_deps/cuco-src/include/cuco/detail/open_addressing/open_addressing_impl.cuh798: cudaErrorInvalidDevice invalid device ordinal
...
Exception in task 0.0 in stage 39.0 (TID 95) ai.rapids.cudf.CudfException: parallel_for failed: cudaErrorInvalidDevice: invalid device ordinal

• row_conversion_test.py::test_row_conversions_fixed_width_wide

ai.rapids.cudf.CudfException: parallel_for failed: cudaErrorInvalidDevice: invalid device ordinal

Steps/Code to reproduce bug
Please provide a list of steps or a code sample to reproduce the issue.
Avoid posting private or sensitive data.

Expected behavior
A clear and concise description of what you expected to happen.

Environment details (please complete the following information)
GB100 + cuda12.8

Revision

spark-rapids: 1b0bce5 (branch-25.02)
spark-rapids-jni: f6fd3f96f286b88d3e6e7bdf2a13dede753b2d09 (branch-25.02)
cudf: 1fe744fb43044e9beb728c244f3f4a0beac3588f (branch-25.02)

Additional context
Add any other context about the problem here.

[pxLi]

cc @ttnghia @sameerz for visibility