Test zonal on geometries with extents that go beyond the raster

[asinghvi17]

We could probably degrade these tests to rasterize. But IMO it's pretty important to test these edge cases.

If these tests pass, we can enable tiled zonal on pretty much any chunk size, so you could calculate the max elevation on Greenland using a 30-meter cell size DEM like Copernicus, all on your local laptop.

This spawned #816.

[rafaqz]

Yeah the edge cases are important, but zonal is very abstract and hardly does anything itself.

Its not actually rasterize that matters either, rasterize is a really complicated set of tools for writing a wide range of things into rasters where they intersect geometries in some way or other, using burn_geometry! underneath.

zonal is just mask, which is boolmask, which internally is also burn_geometry!

Probably testing boolmask more is the cleanest thing. But we could aslo just test burn_geometry! directly. But boolmask should return all trues here, so the test is just all.

(although top level intergration tests like zonal are pretty good too really)

[rafaqz]

Was an edge search optimization breaking things, guess I slightly misunderstood the space when I wrote it or just made a dumb mistake.

I think this is solved with very little additional overhead, we just need to look for off-grid points early in the scan or we lose the start line and don't know we are inside a polygon. Tracking the minimum possible y position does that.

Hopefully there is nothing this bad left around.

(also moved your test to mask as the problem is easier to see there)