Optimize visible coordinates retrieval

[mourner]

Simplifies and significantly optimizes visible coordinates retrieval. A part of my performance push for fixing #6643. Contains two independent changes (please rebase instead of squashing):

• The first commit simplifies and optimizes visible coordinates retrieval during the three painter rendering passes. Instead of getting the same coords multiple times, we retrieve them once for each source at the top and then use the cached objects.
• The second commit fixes a performance regression introduced by sort renderable tiles by y and x to match -native #5601, which made the sorting logic for renderable ids more expensive. The commit makes it only use fancy sorting for symbol layers, and simple zoom sorting for everything else.

Combined, on a test case from #6643, the PR makes retrieving coordinates 3 times faster — on a single fly animation profile, down from 700ms (21.8% of _render) to 230ms (7.8%).

• Revert test changes in sort renderable tiles by y and x to match -native #5601 to make the build pass.

Launch Checklist

• briefly describe the changes in this PR
• write tests for all new functionality already covered
• document any changes to public APIs
• post benchmark scores
• manually test the debug page

[ChrisLoer]

It doesn't look like any behavior changes here, but it got me thinking -- would it be better/more consistent if we drew debug tile boundaries as the union of tiles in all source caches?

[mourner]

We used to draw debug boundaries for all source caches, but I felt it looked very confusing with multiple sources. You can try it but I think it's more useful to have clear, legible boundaries for at least one source.

[ChrisLoer]

👍 These look like very sensible changes to me. I tried playing around with the code locally to look for situations where non-symbol layers might have become dependent on the symbol sorting but couldn't think of anything.

[mourner]

Pushed a fix that makes the tests pass without reverting them. Benchmarks look unaffected, because they don't exercise heavy zooming or small tiles.