[CI] Enable llvm-test-suite on CUDA Linux target

[pvchupin]

• Update reusable workflow with corresponding config
• Add cuda llvm-test-suite run to pre-commit on Linux
• Fix option typo [NFC]

Failing tests disabled at: intel/llvm-test-suite#1052
Issue: #6285

[cperkinsintel]

Getting lots of failures on the CUDA run of the tests. Are there updated headers we should be using?

[pvchupin]

@steffenlarsen has a fix in progress for a few of these problems. I'm going to mark the rest with UNSUPPORTED:cuda and open a separate issue to address these.
Tagging @AerialMantis

[steffenlarsen]

Couple of patches up for failing tests:

• [SYCL][CUDA][HIP][libclc] Rename libclc NonUniform binops to KHR extensions #6279
• [SYCL][CUDA][libclc] Change suld helper functions to return vectors #6280

Additionally, the following patches should fix some problems seen on Windows testing:

• [SYCL][CUDA] Ignore and warn for mem-advise when requirement not met #6281
• [SYCL] Replace std::fill in single_task with a fill command llvm-test-suite#1049
• [SYCL] Ignore backend warning in discard event test llvm-test-suite#1050
• [SYCL] XFAIL Printf tests for CUDA on Windows llvm-test-suite#1051

In local testing on Windows I still see failure in the following tests after the above changes are applied:

  SYCL :: Sampler/unnormalized-clamp-nearest.cpp
  SYCL :: Sampler/unnormalized-clampedge-nearest.cpp
  SYCL :: Sampler/unnormalized-none-nearest.cpp

If they continue to fail after the patches have been merged I suggest we disable these and file an issue.

[pvchupin]

Ok, after merging bunch of changes above and retesting I see the same 3 fails as @steffenlarsen mentioned. I'm going to disable these and open an issue.

[pvchupin]

/verify with intel/llvm-test-suite#1052

[pvchupin]

I realized "/verify with" is useless here because it doesn't run CUDA testing. I actually need to merge that change first and rerun again.