CI: Restore the old test matrix size

[leofang]

Follow-up of #1315. For convenience our test matrix was inflated during the CUDA 13.1 bring-up. But we really don't need to test 3 CUDA versions for each Python version. The old test matrix can be restored to achieve the same goal.

[leofang]

Thinking about it more, I think this is not necessary and actually aligns with what we did when covering CUDA 11/12. Back then, we tested

• previous major.minor (11.8)
• current major.0 (12.0)
• latest current major.minor (12.9)

Now that 13.1 is out, it makes sense to restore this test matrix (12.9, 13.0, 13.1 or future 13.x).