[NO MRG] Test CUDA 12.2 ARM

[jakirkham]

• Include CUDA 12.2 builds from @jameslamb ( Support CUDA 12.2 #5711 )
• Include GPUTreeShap ARM compilation fixes from @robertmaynard ( Correct #20054-D errors found when compiling on ARM gputreeshap#42 )

cc @dantegd @bdice (for vis)

[bdice]

A couple tests failed, in nearest neighbors rbc. @dantegd Do these look flaky or like real errors? https://github.com/rapidsai/cuml/actions/runs/7550203817/job/20556793932?pr=5719#step:7:2209

=========================== short test summary info ============================
FAILED test_nearest_neighbors.py::test_nearest_neighbors_rbc[10000-4-distance_dims2] - assert 61 == 0
 +  where 61 = len(array([0.03404859, 0.05077938, 0.0466364 , 0.03432905, 0.04402662,\n       0.03807753, 0.03931149, 0.05083709, 0.042646...5303, 0.05475897,\n       0.04552774, 0.01488325, 0.04743127, 0.03355303, 0.03736537,\n       0.055514  ], dtype=float32))
FAILED test_nearest_neighbors.py::test_nearest_neighbors_rbc[10000-25-distance_dims2] - assert 3063 == 0
 +  where 3063 = len(array([0.08589244, 0.06178145, 0.09209934, ..., 0.09683452, 0.12032759,\n       0.055514  ], dtype=float32))
= 2 failed, 13253 passed, 6084 skipped, 634 xfailed, 54 xpassed, 2824 warnings in 2008.89s (0:33:28) =

I’ll rerun to see if it is reproducible.

[bdice]

The failures seen here seem unrelated to the GPUTreeShap update, so I am going to request an admin-merge of rapidsai/gputreeshap#42.

[bdice]

The GPUTreeShap fix has been merged. I updated the commit in #5711 to point to the latest gputreeshap 9d32df85f822f186b5fbf53a9a1fa0251d0cd755, so we can close this.

[jakirkham]

Thanks for following up Bradley! 🙏