Http always monitor cdp

[karlseguin]

There's friction between the CDP polling for data and the page polling for data.

This was previously "improved", by making it so that CDP polling would also active the page polling, and making the page somewhat aware that CDP might need polling.

This has been completely reworked.

1 - If there's a page with no CDP (./lightpanda fetch), then the page is free to exit as soon as all work is done
2 - If there's a page with CDP, the page is the only one doing the polling.
3 - If there's no page with CDP, CDP is doing the polling (which is ok, because, without a page, the only data we can get is a CDP message).

This removes the friction and allows the page to poll/timeout smartly, as it has the most context (e.g. information about scheduled tasks).

While this makes the page slightly CDP aware, this PR also removes the page.request_intercepted field, which poorly served a similar purpose. So, as far as CDP-pollution is concerned, it's a wash.

Slightly unrelated, requestAnimationFrame is now automatically placed in the low priority work queue. Unlike setInterval or other background tasks which are inherently repetitive, requestAnimationFrame is repetitive only because the typical use case is for requestAnimationFrame to call requestAnimationFrame. This circumvents our the high -> low priority migration we apply to repeating tasks, meaning that requestAnimationFrame would keep a page from being "done".