Fix polygonize

[rafaqz]

I forgot to PR this months ago.

But it mostly rewites polygonize and uses the uglier but more practically useful exact pixel approach rather than having any angles or curves in the output. It also fixes some bugs and lets you define the contour cutoff yourself.

@skygering I added a performance optimization on coveredby, because it a huge gain for polygonize.

# with extent check in `coveredby`
julia> @btime polygonize(>(0.6), rand(100, 100), minpoints=3);
  2.985 ms (5774 allocations: 2.92 MiB)

# without it
julia> @btime polygonize(>(0.6), rand(100, 100), minpoints=3);
  15.703 ms (5625 allocations: 2.93 MiB)

I'm not sure if that's the way to do it, it should probably go into _line_polygon_process so its used for everything. But, there are cases where not intersecting should return true and others where it should return false depending on the other arguments?

Extent checks are done generically for all predicates now

Oh and this needs tests!!

[skygering]

Exciting optimizations. I don't think the logic is quite there but let me know what you think about my comments.

[asinghvi17]

We should propagate GI.crs(A) to the created geometries, that gets this PR 50% of the way there to full Raster support (the other 50% would probably be an extension on DimensionalData.jl to support dimensional axes).

Edit: implemented in #143