[cuda] device_num.cpp sometimes failing on UR CI

[JackAKirk]

I have not found examples of this failing on intel/llvm CI, but it is happening on UR CI for apparently unrelated PRs. see e.g.

• https://github.com/oneapi-src/unified-runtime/actions/runs/10767363770/job/29855526054?pr=2069
• https://github.com/oneapi-src/unified-runtime/actions/runs/10848734805/job/30107943832
• https://github.com/oneapi-src/unified-runtime/pull/1979

Maybe it is a general problem and it is just by chance I only see it happening on on UR ci. I note that l0 currently has the test xfail'd due to flaky failures. Or maybe it is something specific to UR ci.