In a fast debug build from treeherder, this testcase always hits the assertion. But with a slow local debug build, I have to hit the back button several times to trigger the assertion.

What is happening here is that we are suspending the AudioContext because `nsGlobalWindow::SuspendTimeouts` is called. At this point, we switch from the `AudioCallbackDriver` to a `SystemClockDriver` because we don't need an audio thread anymore and we'll pause the graph anyways. Then sometime later before the audio channel is closed, and before the promise for `Suspend` is sent back to the content, `nsGlobalWindow::FreeInnerObjects` is called, which calls "Close" on the context. This goes down the usual route, and finds a `SystemClockDriver` that is not switching, with no audio tracks, so it's resolved immediately. This is not very important, and the solution is not trivial (at least I could not immediately find a trivial one), I'm tempted to just relax the assert. Maybe a good way would be to not pass down a promise for Resume/Suspend/Close called from Gecko itself ?

Mass change P3->P4 to align with new Mozilla triage process.

No longer reproduces on trunk.

Pushed by ryanvm@gmail.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/e1d8d4d15bd2 Add crashtest. r=me