DX12 per-slot fence counters signalled into a single shared fence

[CanReader]

src/Graphics/DirectX12/DirectX12Renderer.cpp:776-777, wait loop at :631-642;
writes at :1265 and src/Graphics/DirectX12/DirectX12Buffer.cpp:382

fenceValues[frameIndex]++;
Signal(fence, fenceValues[frameIndex]);

With FrameCount == 2 and both slots starting at 0, the two per-slot counters advance independently but are
signalled into one shared fence. Odd frames therefore re-signal a value the fence has already reached.

The "wait for ALL pending GPU work" loop at :631-642: whose own comment says it exists specifically to
prevent constant-buffer races: consequently short-circuits on every second frame, while the CPU overwrites
mapped UPLOAD-heap memory at :1265 and DirectX12Buffer.cpp:382.

Severity: corruption (races on every other frame).

Found by a full audit of the graphics layer. Assessed severity: corruption: every frame.