NativeEngine: accept R8 and RG32F in raw texture upload path.

[bkaradzic-microsoft]

What

Adds R8 and RG32F to the small format whitelist asserted by PrepareImage in Plugins/NativeEngine/Source/NativeEngine.cpp. Both formats are already supported by bimg and bgfx; the gap was only in BN's adapter assert.

             assert(
+                image->m_format == bimg::TextureFormat::R8 ||
                 image->m_format == bimg::TextureFormat::R16 ||
                 image->m_format == bimg::TextureFormat::RGB8 ||
                 image->m_format == bimg::TextureFormat::RGBA8 ||
                 image->m_format == bimg::TextureFormat::RGBA16 ||
                 image->m_format == bimg::TextureFormat::RGBA16F ||
                 image->m_format == bimg::TextureFormat::RG16F ||
+                image->m_format == bimg::TextureFormat::RG32F ||
                 image->m_format == bimg::TextureFormat::RGBA32F ||
                 image->m_format == bimg::TextureFormat::RGBA32U);

Why

NativeEngine::LoadRawTexture (the binding for Babylon.js engine.createRawTexture(...)) forwards the JS-supplied texture format to PrepareImage, which had a hard-coded whitelist of 8 formats. Two formats that Babylon.js requests in normal usage fall outside that list and cause SIGABRT on the assert:

Format	Where it shows up
R8	createRawTexture for single-channel byte data (texture updates)
RG32F	GPU particle internal state textures (262 elements * 8 bytes)

RG16F was already on the list - the 32-bit twin had simply been overlooked.

How verified

Local Win32 V8 D3D11 Debug build, headless Playground runner, 8 currently-excluded tests known to hit this assert:

idx	title	before	after
336	Test updateTextureData	SIGABRT (R8)	runs (hits separate JS issue, not in scope)
652	Particles - Billboard Y	SIGABRT (RG32F)	runs (pixel-diff fail)
682	GPU Particles - Change - Size	SIGABRT (RG32F)	runs (pixel-diff fail)
684	GPU Particles - Change - Speed	SIGABRT (RG32F)	runs (pixel-diff fail)
685	GPU Particles - Change - Speed Limit	SIGABRT (RG32F)	runs (pixel-diff fail)
686	GPU Particles - Change - Angular Speed	SIGABRT (RG32F)	runs (pixel-diff fail)
687	GPU Particles - Change - Drag	SIGABRT (RG32F)	runs (pixel-diff fail)
688	GPU Particles - Change - Size Range	SIGABRT (RG32F)	runs (pixel-diff fail)

Each failing call was confirmed by a temporary fprintf placed in LoadRawTexture (build, capture, revert).

Scope

• This PR only fixes the assert. The 8 tests still fail and remain excluded - whether they should be re-enabled is a separate question, since they fail pixel-diff (and idx 336 surfaces an unrelated JS issue) after the assert is removed.
• For these tests mips=false and srgb=false, so the mipmap-generation and SRGB-conversion branches in PrepareImage don't need format-aware extensions for this fix. If a future caller requests mips on R8 / RG32F, the existing mip branch will need analogous handling.
• FlipImage is format-agnostic (it computes rowPitch from total-size / height), so invertY is safe for the new formats.

Risk

Very low. Two enum values added to a disjunction. No code path that previously rejected these formats now accepts data that bgfx would refuse to ingest - the underlying renderer already advertises both formats.

[Copilot]

Pull request overview

This PR updates the NativeEngine raw texture upload preparation path to accept two additional bimg texture formats (R8 and RG32F) that are already supported downstream (bimg/bgfx), preventing debug-time aborts when Babylon.js requests these formats via engine.createRawTexture(...).

Changes:

• Extend PrepareImage’s debug-format whitelist to include bimg::TextureFormat::R8.
• Extend PrepareImage’s debug-format whitelist to include bimg::TextureFormat::RG32F.