D3D9On12: pass nullptr to transient ResidencySet::Open

[vdwtanner]

Problem

Consumer-side update to track the breaking signature change for D3D12TranslationLayer::ResidencySet::Open() introduced in:

• Fix dangling-pointer walk in ResidencySet during DiscardCommandList D3D12TranslationLayer#114

See that PR for full root-cause analysis.

Change

D3D9on12::Device::UnwrapResource (9on12Device.cpp:308) constructs a transient ResidencySet, opens it on the MAX_VALID (transient) command-list slot, inserts one residency handle, closes, and submits — all in a single call frame on the same thread. The set is never visible to EndTrackingObject, so manager registration is pure overhead.

Passing nullptr for the new ResidencyManager* parameter opts this call site into the "no tracking, no locking" mode, matching its actual lifetime.

-            pResidencySet->Open((UINT)D3D12TranslationLayer::COMMAND_LIST_TYPE::MAX_VALID);

+            // Transient set: lifetime is bounded to this call frame, so it cannot race with a destroyed

+            // Resource via ResidencyManager::EndTrackingObject. Pass nullptr to skip manager registration.

+            pResidencySet->Open((UINT)D3D12TranslationLayer::COMMAND_LIST_TYPE::MAX_VALID, nullptr);

Dependencies

• Fix dangling-pointer walk in ResidencySet during DiscardCommandList D3D12TranslationLayer#114 must merge first (defines the new 2-arg signature). This PR's CI will not pass against the current D3D12TranslationLayer:master until that lands and the submodule pointer here is bumped.

Verification

Compile-clean build of d3d9on12.dll (Debug x64) against the updated D3D12TranslationLayer branch. Same d3d9on12.dll was used for the end-to-end TAEF repro in the D3D12TranslationLayer PR's verification section (the test that exercises the underlying UAF runs through this DLL).

Pre-empted review questions

• Behavior change at this call site? None — the transient set never had a real concurrent-destroy exposure; nullptr just removes the overhead the new tracking would otherwise add.

• Regression risk? None — mechanical 1-line change, same lock-free fast path as before.

[vdwtanner]

Closing -- D3D12TranslationLayer PR #114 pivoted to a no-API-change design per Jesse's review (microsoft/D3D12TranslationLayer#114 (comment)), so no consumer-side update is needed. The dependency this PR existed to track no longer exists.