Removed all D3D12 exclusions from tests.

[bkaradzic-microsoft]

Re-enables the scissor test cases that were previously excluded on D3D12 due to an automatic mipmap generation issue in bgfx.

[Copilot]

Pull request overview

Re-enables the previously D3D12-excluded scissor visual tests by updating bgfx integration and adjusting framebuffer attachment resolve behavior to avoid newer bgfx validation asserts around auto-mip generation.

Changes:

• Update bgfx.cmake FetchContent tag to a newer commit.
• Adjust framebuffer attachment initialization to conditionally enable BGFX_RESOLVE_AUTO_GEN_MIPS only when the texture format supports mip autogen, and explicitly disable resolve for depth attachments.
• Remove the D3D12 exclusions (and related comments) for scissor tests in the playground validation config.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
Plugins/NativeEngine/Source/NativeEngine.cpp	Guards auto-mip resolve flag based on caps to prevent bgfx validation asserts; explicitly disables resolve for depth attachment.
CMakeLists.txt	Bumps bgfx.cmake dependency commit used by the build.
Apps/Playground/Scripts/config.json	Re-enables scissor tests on D3D12 by removing exclusions.

[bghgary]

[Review by Copilot on behalf of @bghgary]

Nice fix, and the comment on the color attachment in CreateFrameBuffer captures the subtlety well. One concern: there are other Attachment::init(...) call sites using the short overload that rely on the same default-ignores-unsupported-formats behavior and will hit the new bgfx assert after this bump. They aren't covered by the scissor tests, so CI going green won't prove they're safe:

• Polyfills/Canvas/Source/FrameBufferPool.cpp:65 — loops over a color (RGBA8) and a depth (D24S8) attachment, both with the default overload. The depth one is the concern: depth formats don't support auto-mipgen, so this should reliably trip the new assert once Canvas is exercised.
• Plugins/NativeXr/Source/NativeXrImpl.cpp:285-286 — same pattern, color + depth, both relying on defaults.

Could we fix these in the same PR? The change is mechanical (query caps->formats[...] & BGFX_CAPS_FORMAT_TEXTURE_MIP_AUTOGEN for color, explicit BGFX_RESOLVE_NONE for depth), and leaving them for later means a known assert waiting to fire on the first Canvas/WebXR workload after the merge.