Change DOM surveyor to time-based processing logic (from chunk-based)

The DOM surveyor will now use time-based logic to spread its work
over time. This allows the surveying to better scale down on
slower devices.

Additionally, the DOM surveyor code has been reworked to lower as
much as possible memory churning when collating nodes to survey.

This rework has been motivated after profiling the "monstrous DOM"
seen in the following page:
<https://doc.rust-lang.org/std/iter/trait.Iterator.html>

The idea is that making the DOM surveyor efficient on such
"monstrous DOM" case should make it efficient everywhere in
practice.