bug: Fix SpatialFilter for correct pruning when reading from geoparquet

[petern48]

This PR fixes some bugs in the SpatialFilter used for pruning. The code now follows the same logic as used in the original Apache Sedona code here. Essentially, it uses a Covers SpatialFilter instead of a CoveredBy, and updates the code accordingly.

For example, st_covered_by function was modified in the following way

• (old logic) "if (column, literal), use a CoveredBy filter, else if (literal, column, use a Intersects filter."
• (new logic) "if (column, literal), use an Intersects filter, else if (literal column, use a Covers filter."

closes #49

[petern48]

Hi, it's me 🤠. I originally picked this up thinking this would be a simple one-liner fix. However, when I ran all of the predicates against geopandas output (the integration test), I saw there were more buggy predicate functions (e.g st_covers, st_contains, st_coveredby, st_within) than the original ones I expected.

Using new integration tests, I specifically found that the (column, literal) logic for coveredby was buggy (e.g CI failure). I honestly didn't think too hard about why that logic is buggy, but I figured it's easier and apparently more correct to switch to the Covers filter logic the original Apache Sedona code uses.

You're welcome to propose a different direction, if you can think of one.

[Kontinuation]

Yes, the fix is correct.

Only contains/covers can be pushed down, not within/covered_by.