Optimize tile calculations for globe

[mourner]

When profiling globe view, I noticed that aabbForTileOnGlobe takes ~5% of CPU time, and makes quite a lot of allocations for a hot path function that's called many times per frame. This PR eliminates most redundant allocations, bringing it below 1% in the profiler. See the two self-contained commits:

1. Instead of allocating a new scaled globe matrix every time, use transform's cached globeMatrix and scale things inline when calculating bounding volume corners.
2. Eliminate most temporary arrays that are immediately discarded. As a part of this, also clean up some function signatures so that LngLat corners of a bounding box are encapsulated in LngLatBounds which we designated for such cases.

Launch Checklist

• briefly describe the changes in this PR
• manually test the debug page
• tagged @mapbox/gl-native if this PR includes shader changes or needs a native port
• apply changelog label ('bug', 'feature', 'docs', etc) or use the label 'skip changelog'
• add an entry inside this element for inclusion in the mapbox-gl-js changelog: <changelog>Improve rendering performance on globe view</changelog>

[mourner]

@akoylasar the first commit could be ported to native since it appears to have the same redundancy.

[karimnaaji]

👍