Remove unused pWaitDstStageMask from Vulkan submit

[MarijnS95]

pWaitDstStageMask is only meaningful when wait semaphores are used, which this submit path does not use. This wasn't causing errors in practice because vkWaitForFences() is called after every submit, preventing concurrent in-flight command buffers.

The compute dispatch submit passed VK_PIPELINE_STAGE_TRANSFER_BIT, intending to wait for the prior buffer transfer submit to complete, but pWaitDstStageMask is a destination stage mask — it specifies which stages in the current submit to block until a wait semaphore signals, not which prior stages to wait on. Without semaphores this had no effect.

[manon-traverse]

LGTM

[MarijnS95]

Let my double-check why this suddenly "fixes" some tests on the Windows GPU:

Unexpectedly Passed Tests (3):
  OffloadTest-clang-vk :: Feature/StructuredBuffer/matrix.test
  OffloadTest-clang-vk :: Feature/StructuredBuffer/matrix_assign.test
  OffloadTest-clang-vk :: Feature/Textures/Texture2D.OperatorIndex.test.yaml

[bogner]

Summary

• Remove pWaitDstStageMask assignment from executeCommandBuffer since no wait semaphores are used
• Remove the unused WaitMask parameter and the VK_PIPELINE_STAGE_TRANSFER_BIT argument at the call site

See commit message for details on why this was harmless but incorrect.

When putting up a PR the description should be the commit message. The git history during review inevitably collects various fixups, and burying the intended commit message there makes it hard to review the message itself.

[MarijnS95]

@bogner you're right, I've configured Claude to not have the useless # Summary header especially if that's the only section, and to keep the commit message verbatim (GitHub's default anyway) for single-commit PR's, yet it still seems to create these meaningless rewrites of my original message (although I should have checked the final thing it put up). Going back to good-old gh pr create next time :)