[GH-3329] Preserve WKB dimensions in Arrow export

[ryux1]

Did you read the Contributor Guide?

The applicable contributor rules and Python development instructions were checked during delegated preparation.

Is this PR related to a ticket?

Yes. Closes #3329.

What changes were proposed in this PR?

For WKB encoding, collect the canonical output of GeoSeries.to_wkb() before delegating Arrow construction to GeoPandas. This avoids a GeometryType-to-Python round trip that can infer a Z dimension for a 2D GeometryCollection whose typed members are all empty. Native GeoArrow encoding continues to use the existing path.

The shared parity fixture now includes GeometryCollection([Point(), LineString(), Polygon()]), reproducing the case removed from #3326 and verifying that Sedona matches GeoPandas.

How was this patch tested?

• Python 3.9 with PySpark 3.5 and the Sedona 1.9.0 shaded JAR
• TestMatchGeopandasSeries::test_to_arrow
• TestMatchGeopandasSeries::test_to_wkb
• TestGeoSeries::test_to_arrow
• all configured pre-commit hooks for both changed files
• Python bytecode compilation and git diff --check

Did this PR include necessary documentation updates?

No. This fixes serialization behavior without changing the public API.

[jiayuasu]

Thanks for creating this PR. the correct fix should be in this PR #3335

[ryux1]

Thanks for pointing me to #3335. I checked the two approaches, and #3335 fixes the dimension loss in the shared decoder rather than special-casing WKB export, with broader empty LineString and GEOS-version coverage. It fully supersedes this narrower change, so I’m closing this PR in favor of it.