[SYCL][OpenCL] Ban AMD OpenCL platform (#5825)

[al42and]

Compiling kernels for it is not supported, device info sometimes throws, and there is the HIP backend for those wanting to use an AMD GPU.

$ sycl-ls  # before
[opencl:gpu:0] Intel(R) OpenCL HD Graphics, Intel(R) UHD Graphics 770 [0x4680] 3.0 [22.35.24055]
[opencl:gpu:1] AMD Accelerated Parallel Processing, gfx1032 2.0 [3452.0 (HSA1.1,LC)]
[ext_oneapi_level_zero:gpu:0] Intel(R) Level-Zero, Intel(R) UHD Graphics 770 [0x4680] 1.3 [1.3.24055]
[ext_oneapi_cuda:gpu:0] NVIDIA CUDA BACKEND, NVIDIA GeForce RTX 3060 0.0 [CUDA 11.7]
[ext_oneapi_hip:gpu:0] AMD HIP BACKEND, gfx1032 0.0 [HIP 50221.15]

$ sycl-ls  # after
[opencl:gpu:0] Intel(R) OpenCL HD Graphics, Intel(R) UHD Graphics 770 [0x4680] 3.0 [22.35.24055]
[ext_oneapi_level_zero:gpu:0] Intel(R) Level-Zero, Intel(R) UHD Graphics 770 [0x4680] 1.3 [1.3.24055]
[ext_oneapi_cuda:gpu:0] NVIDIA CUDA BACKEND, NVIDIA GeForce RTX 3060 0.0 [CUDA 11.7]
[ext_oneapi_hip:gpu:0] AMD HIP BACKEND, gfx1032 0.0 [HIP 50221.15]

Closes #5825

[v-klochkov]

The change looks good to me, but I may not know some plans regarding supported platforms, etc.
@pvchupin can you please assign someone from your team to review this PR?

[steffenlarsen]

If OpenCL for AMD GPUs is being problematic, I am okay with giving it the same treatment as we have done for NVIDIA GPUs. @AerialMantis - Is this something you have tested and/or had problems with before?

[al42and]

May I bump this? OpenCL backend still does not support AMD GPUs as far as I can tell, and listing them among available devices is just confusing.

[npmiller]

Yes, as far as I'm aware this is also not working, like the Nvidia one, I'm happy to ban it as well for now.

[bader]

@al42and, could you update your branch, please? I see that it was tested in September last year, I'd like to make sure it still passes the tests.
@steffenlarsen, @v-klochkov, could you formally approve the patch, please?

On comment from my side, I think we should discard devices based on features they do not support (e.g. required SPIR-V extension) instead of relying on platform name. This suggestion is better implement in follow-up patches.

[al42and]

@al42and, could you update your branch, please? I see that it was tested in September last year, I'd like to make sure it still passes the tests.

Rebased on top of the current sycl branch.

[al42and]

Failed Tests (1):
    SYCL :: XPTI/kernel/basic.cpp

There are other failing GitHub Actions with the same error, so I guess that's not this PRs fault

[bader]

Failed Tests (1):
    SYCL :: XPTI/kernel/basic.cpp

There are other failing GitHub Actions with the same error, so I guess that's not this PRs fault

This issue was caused by 18899cc, which was reverted in 01511a3.

[bader]

On comment from my side, I think we should discard devices based on features they do not support (e.g. required SPIR-V extension) instead of relying on platform name. This suggestion is better implement in follow-up patches.

In addition to that it can be useful to list discarded devices with --verbose flag with a clear mark + a comment about why were discarded.