vulkan: add fwht support for Intel with shmem reduction

[0cc4m]

Overview

Add a FWHT shader path that does not rely on subgroup size and collectives for Intel GPUs.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES, Claude wrote the code, I corrected and reviewed.

[0cc4m]

I don't get the failure of the CI Windows Intel run, on Linux. @rillomas Any ideas?

[jeffbolznv]

I don't get the failure of the CI Windows Intel run, on Linux. @rillomas Any ideas?

I think the workgroup size might be too large, 512*4 (would be above NVIDIA's limit, at least). Check VVL?

[0cc4m]

Right, I missed that. I'll just keep it closer to the subgroup-based shader, that should also be easier to maintain.

[rillomas]

I don't get the failure of the CI Windows Intel run, on Linux. @rillomas Any ideas?

The crash in Windows CI was the same as #23687 (comment) . I also checked the latest driver 32.0.101.8826 but it was still crashing so we'll need to wait until the fix is deployed, at least for Xe2/3 environments.