Add a test for raster overlay upsampled bounding volumes, and fix a bug with upsampling and gltfUpAxis
This is a follow on to #917 that adds a test for one of the bug fixes in that PR:
Fixed a bug while upsampling tiles for raster overlays that could cause them to have an incorrect bounding box, which in some cases would lead to the raster overlay being missing entirely from the upsampled tile.
However, in the course of writing this test, I found an unrelated bug where raster overlay upsampling was not correctly accounting for the non-standard but relatively widely used gltfUpAxis property. So this PR fixes that as well.
I'm tempted to merge this myself to include in tomorrow's release, since it's a useful fix and low risk. But I'll move it out of the release instead so that it can have a proper review.
Thanks @j9liu, not sure how I missed that test failure. But it's fixed and this is ready for another look.
Thanks @kring! Merging now.