ci: run the Vulkan game tests with synchronization validation

[sasvdw]

PR Details

Follow-up to #3324, which adds the STRIDE_VULKAN_SYNC_VALIDATION=1 switch. This sets it on the Vulkan leg of test-windows-game.yml, so a missing barrier fails the test instead of passing silently.

No new matrix leg and no callback change. GameTestBase.RunGameTest already collects every DebugLogModule error during a run and fails the test with them; synchronization hazards arrive at error severity. The switch is inert until #3324 merges, and it is inert on the Direct3D legs.

Every suite in build\Stride.Tests.Game.GPU.slnf was run on Vulkan (Lavapipe, Debug) with the switch on the #3324 head: Stride.Graphics.Tests, Stride.Graphics.Tests.10_0, Stride.Graphics.Tests.11_0, Stride.Engine.Tests, Stride.Physics.Tests and Stride.UI.Tests all pass with no hazard reported. Without #3324's access-mask fix, TestHammersley and TestRadiancePrefilteringGgxFaceContinuity fail on READ_AFTER_WRITE at vkCmdBeginRendering, which is the failure this leg exists to catch.

Related Issue

Follow-up to #3324.

Types of changes

• Docs change / refactoring / dependency upgrade
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist

• My change requires a change to the documentation.
• I have added tests to cover my changes.
• All new and existing tests passed.
• I have built and run the editor to try this change out.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VHYKv129fEUp6sxMbybZWf

[xen2]

Tested on my fork (this commit on top of current master, which has #3324):

• Full Vulkan game tests with the switch on: 446 passed, no hazard, Test step 2m01 (normal runs: 1m55 to 2m06, so no measurable cost). https://github.com/xen2/stride/actions/runs/35524227620

I also ran a canary to confirm the switch is really active on CI and adds coverage. I removed the two attachment read bits in BarrierMapping.ToVkAccessFlags and ran TestRadiancePrefilteringGgxFaceContinuity:

• Switch on: fails with 11 READ_AFTER_WRITE errors. https://github.com/xen2/stride/actions/runs/35549768147
• Switch off (same code, env line removed): passes. https://github.com/xen2/stride/actions/runs/35549769475

So core validation alone does not catch it, and this change does. Note the checks on this PR skipped all test jobs (workflow-only change), so the runs above are the actual test. Good to merge.

Follow-up: I also tried the switch on the sample lanes (test-enduser). It found two real barrier bugs around the read-only depth buffer that no unit test reaches (a storeOp on a read-only depth, and DepthStencilRead not covering the shader read used by soft particles).
Both are fixed and green on my fork, and I will open a separate PR for the fixes plus the switch on the samples and maybe editor as well.