Servo enters death spiral if frontend outpaces webrender backend

[Gankra]

If you simulate the webrender backend struggling by adding the following to the ApiMsg::SetDisplayList handler in render_backend.rs:

::std::thread::sleep(::std::time::Duration::from_millis(20));

servo enters a death spiral where it just endlessly piles display lists into the queue for webrender to render. Each frame gains ever-increasing latency, and any attempt to interact with the page (i.e. scrolling) triggers a complete lock up, as the system blocks on catching up.

You can visibly see this with -Zwr-stats, as the "message send time" begins to monotonically climb (message send time being time_display_list_processed - time_display_list_built).

In more realistic situations where we only falter for a few frames, this causes perceptible lag in interactions.