Optimize and simplify line tessellation code

[mourner]

This PR refactors the line tessellation code, removing code repetition and some unnecessary logic, making it both faster and easier to follow.

The potential performance improvement comes from reducing the amount of temporary Point object allocations by ~2 times, which theoretically reduces the amount of subsequent GC pauses. I'll try to find a way to measure real-world performance impact, but roughly it should be 5-10% less time spent tessellating lines on the worker side.

It brings some ideas from #7085 but in a less intrusive way.

The PR is made on top of #8275 so needs to be applied after it lands.

Launch Checklist

• briefly describe the changes in this PR
• write tests for all new functionality already covered
• document any changes to public APIs no changes
• post benchmark scores
• manually test the debug page
• tagged @mapbox/gl-native — might need a port to native to make the code between platforms more similar.

[ansis]

This looks great! just that one small question.

It's amazing how much unnecessary stuff was in there. This should be much easier to understand.

[ansis]

Do you think creating a line for each object is a significant cost? putting per-line values directly on the bucket seems a bit less nice but probably ok if there is a significant benefit.

[mourner]

@ansis we had distance as a bucket state before this PR, so the latter doesn't change the status quo in that sense — I moved the line stats to shared state too for simplicity/consistency, but I think we can revisit this in a follow-up PR.

[mourner]

Layout benchmarks are not affected: