[BingoWings247]

Goal of this PR

Fix two defects in the application/x-cfx-game-view compatibility renderer (EpoxyScript.h) that keep a full-frame WebGL2 renderer running for every game-view element a page has ever mounted (#4342):

1. The only teardown path is canvas.addEventListener('remove', ...). remove is not a DOM event, so it never fires when a page removes the element; the requestAnimationFrame loop and the WebGL2 context (powerPreference: 'high-performance') stay alive.
2. CfxGameViewRenderer.destroy() calls this.#canvas.parentNode.removeChild(canvas), where canvas is not defined in that scope. Whenever the canvas still has a parent (including a container that was itself detached), destroy() throws ReferenceError: canvas is not defined after cancelling the rAF loop but before WEBGL_lose_context, so the context is never released and the listener never reaches resizeObserver.disconnect(). This also breaks resources that work around (1) by dispatching remove themselves.

How is this PR achieving the goal

• destroy(): removeChild(canvas) → removeChild(this.#canvas).
• The compat layer now tracks the canvases it created a renderer for (LiveCanvases). The existing MutationObserver, which already watches the whole document, checks them on each mutation batch. A canvas that is no longer isConnected gets a 1 second grace period, so moving it between containers or briefly detaching it keeps its renderer; after that the existing teardown runs by dispatching the remove event the listener was already waiting for. The listener is now { once: true } and removes the canvas from the set.
• The check only runs while at least one compat renderer is alive and is O(live renderers).
• Renderers created directly with new CfxGameViewRenderer(canvas) are unaffected (their owner calls destroy()), and resources that already dispatch remove keep working.
• Known trade-off: a compat canvas that is re-inserted more than 1 second after being removed stays blank, the same as any destroyed renderer.

Before/after, using the g_gameViewScript extracted from master 88a4138 and injected into a page the way NUI does:

Scenario	Before	After
Mount and unmount a container 10 times	10 render loops running, 10 WebGL2 contexts alive	0 loops, all 10 contexts released
Remove the canvas element itself	render loop keeps running	released after the grace period
Page dispatches remove while the canvas is attached	ReferenceError: canvas is not defined, context stays alive	released, canvas removed
Move the canvas to another container in the same task	keeps rendering	keeps rendering
Detach, then re-attach after 300 ms	keeps rendering	keeps rendering
Element stays mounted (control)	1 loop	1 loop

This PR applies to the following area(s)

FiveM, RedM

Successfully tested on

Game builds: n/a. The change is JavaScript inside the g_gameViewScript raw string literal; I verified the literal still terminates in the same place and the extracted script parses.

Platforms: Headless Chromium (ANGLE). Not built in a client locally; CI will compile it.

Checklist

• Code compiles and has been tested successfully.
• Code explains itself well and/or is documented.
• My commit message explains what the changes do and what they are for.
• No extra compilation warnings are added by these changes.

Fixes issues

fixes #4342

[BingoWings247]

The Windows build failures are from the CI runner, not this change: .github/workflows/ci/build_windows.sh
calls MSBuild at C:\Program Files\Microsoft Visual Studio\2022\Enterprise\MSBuild\Current\Bin\MsBuild.exe,
which isn't present on the current windows-latest image (exit 127 before compilation starts).
Any PR will hit this until the path or runner image is updated.