Throw from Device::UpdateDevice when rendering is enabled

[bghgary]

Context

Babylon::Graphics::Device::UpdateDevice(DeviceT) silently no-ops when called at the wrong time. If it's called while bgfx is initialized (after the first EnableRendering / StartRenderingCurrentFrame), the new device pointer gets stored in Bgfx.InitState.platformData and Dirty = true, but the swap does not actually take effect. bgfx::init only runs again on the next EnableRendering, which early-outs while Bgfx.Initialized == true.

Inside DeviceImpl::Frame(), UpdateBgfxState() runs each frame and, when Dirty, calls bgfx::setPlatformData + bgfx::frame(DISCARD) + bgfx::reset. So the new device pointer reaches g_platformData.context, but the bgfx renderer's m_device is captured at bgfx::init time and is never re-read by bgfx::reset on any backend (D3D11, D3D12, Metal, OpenGL). The caller's swap appears to succeed but the device never actually changes.

#1687 adds a TEST(Device, UpdateDevice) that exercises the canonical correct flow (DisableRendering → UpdateDevice → frame); this PR makes the wrong flow fail loudly instead of silently.

Change

• DeviceImpl::UpdateDevice throws std::runtime_error when called while m_state.Bgfx.Initialized == true.
• Device::UpdateDevice doc-comment in Device.h states the contract: only valid when rendering is disabled; throws otherwise.
• New test TEST(Device, UpdateDeviceThrowsWhenRenderingEnabled) in Tests.Device.D3D11.cpp. Verifies pre-init UpdateDevice is permitted, post-init throws, and the swap completes after DisableRendering.

Why only UpdateDevice and not the rest of the Update* family?

The Update* methods fall into two groups based on whether bgfx::reset (called from UpdateBgfxState) actually applies the change:

Method	Applied via bgfx::reset?	Notes
UpdateSize	Yes	bgfx::reset's primary purpose is resolution change
UpdateMSAA	Yes	BGFX_RESET_MSAA_* flags consumed by reset
UpdateAlphaPremultiplied	Yes	BGFX_RESET_TRANSPARENT_BACKBUFFER flag
UpdateBackBuffer (D3D11)	Yes	g_platformData.backBuffer re-read by reset
UpdateWindow	Backend-dependent — being addressed separately in bgfx
UpdateDevice	No on every backend	renderer's m_device AddRef'd at init, never swapped by reset

UpdateDevice is uniquely "init-only on every backend" — making it throw is a clean cut. UpdateWindow has analogous inconsistency across bgfx backends and is being addressed in bgfx itself.

Local verification

• D3D11 full UnitTests: PASSED including the new Device.UpdateDeviceThrowsWhenRenderingEnabled.
• D3D12 builds; existing tests unaffected (no D3D12-specific UpdateDevice test in this PR — Add Device.UpdateDevice unit test for D3D11/D3D12 #1687 covers that).

[Created by Copilot on behalf of @bghgary]

[Copilot]

Pull request overview

This PR makes Babylon::Graphics::Device::UpdateDevice fail loudly when called after bgfx has been initialized (i.e., while rendering is enabled), avoiding a silent no-op device “swap” that never actually takes effect on any backend.

Changes:

• Throw std::runtime_error from DeviceImpl::UpdateDevice when bgfx is already initialized.
• Document the UpdateDevice contract in Device.h (only valid while rendering is disabled).
• Add a D3D11 unit test asserting UpdateDevice throws when rendering is enabled.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
Core/Graphics/Source/DeviceImpl.cpp	Adds a guard that throws if UpdateDevice is called after bgfx initialization.
Core/Graphics/Include/Shared/Babylon/Graphics/Device.h	Documents that UpdateDevice is only valid when rendering is disabled and will throw otherwise.
Apps/UnitTests/Source/Tests.Device.D3D11.cpp	Adds a new D3D11 unit test covering the “throws when rendering enabled” behavior.