fix: avoid wasm panic without GL context (WebGL2 only)

[pietrosophya]

Objective

• Partially fixes WASM: let the app work if the GL context is lost #12681
• Partially fixes Bevy should recover from WebGL Context Lost or WebGPU Device Lost #10456

Solution

If the context lost is detected (checking for the canvas, then for the WebGL2 rendering context, then for the is_context_lost()), it pauses the subapps.

This has the effect of avoid calling GL related operations, thus letting the rest of the app work.

I don't know:

• if the pausing/resuming mechanism can be replaced by something else already existing
• if only specific subapps (e.g. the RenderApp) can be paused

This fixes only WebGL2 because there's yet no support for WebGPU to check context lost (there's a promise handler for Chrome but it's still an experimental feature).

[pietrosophya]

I checked if we can at least retrieve the GpuCanvasContext for WebGPU, but this is still unstable and would require a special config when running the project.

[alice-i-cecile]

@maximetinu @beoboo if either of you have relevant expertise here, a review would be very welcome. Even validating that this does fix your issue would be great.

[pietrosophya]

Yep, it does fix our issue. since it disables the render app if there is no GL available (and thus avoids a crash).

[pietrosophya]

I fixed all conflicts.

[maximetinu]

if either of you have relevant expertise here, a review would be very welcome. Even validating that this does fix your issue would be great.

Sorry for the delayed review @alice-i-cecile

So, on one hand, this does fix the issue because Bevy doesn't panic anymore when it loses the webgl2 context.

But, on another hand, I don't think this is a general-enough solution worth contributing upstream for a couple of reasons:

• it pauses not only the render sub-app, but all the sub apps. Other Bevy users could be defining their own sub app for some other non-graphics related thing, and they may not want to pause it when the webgl2 context is lost.
• ideally, we should just pause the render app, but this would introduce a coupling between bevy_winit and bevy_render
• so, in reality, pausing all the sub apps for this reason is introducing a hidden coupling between bevy_winit and bevy_render anyway, but preventing a graphic crash like this should be responsibility only of bevy_render (if not of the wgpu crate itself, ideally).

I think that a good solution would require:

• allow pausing only specific sub apps
• do this check as part of bevy_render, so it pauses itself

I have discussed this offline with @pietrosophya, and he has an idea to allow each app or sub app to implement a customizable fn should_update(..) -> bool system that checks for pre-conditions like this. In the case of the RenderApp sub app, in wasm+webgl2, it'd check that the webgl2 context still exists / has not been lost. That sounds very good to me, but it's a bigger feature out of scope for this PR

[pietrosophya]

After thinking more about this, and after trying to introduce a filter for SubApps, I think this remains a good solution for now, since the crash can happen in any app that contains the RenderApp (maybe 99% of the apps out there), so IMO it's ok to be merged as it is now.