GeoPandas: implement distributed coverage validation

[jiayuasu]

Part of #2230.

GeoPandas 1.1 added GeoSeries.is_valid_coverage and
GeoSeries.invalid_coverage_edges for validating polygonal coverages. A valid coverage has
non-overlapping polygon interiors and exactly edge-matched shared boundaries; callers may also
request detection of gaps up to a specified width.

Sedona already depends on JTS 1.20, which provides CoveragePolygonValidator, but validating an
entire distributed GeoSeries cannot be implemented by collecting every polygon into one task.

The implementation should:

• add a focused internal Spark primitive that validates one target geometry against a
  self-including array of candidate coverage geometries;
• discover candidate neighbors with a distributed spatial self-join using polygonal envelopes
  expanded by gap_width, matching JTS coverage-index semantics;
• validate each distinct represented geometry against its candidates on executors, then map the
  diagnostic back to every matching physical input row;
• return one invalid-edge geometry per original row while preserving its index, order, CRS, and
  output naming;
• reduce the distributed invalid-edge result to one Python boolean for is_valid_coverage;
• preserve GeoPandas behavior for valid coverages, overlaps, edge mismatches, duplicate polygons,
  narrow gaps, holes, multipolygons, empty sources, and empty/non-polygonal values;
• preserve null geometries in their physical input rows instead of reproducing Shapely 2.1's
  result-compaction bug, which moves null outputs to the end;
• accept finite, non-negative numeric gap_width scalars and reject negative or non-finite values,
  whose behavior is inconsistent between GEOS and JTS (and for positive infinity would require a
  non-scalable all-pairs join); and
• avoid Python UDFs, global geometry collection, and driver materialization of geometry rows.

The per-target primitive is implementation plumbing for the distributed GeoPandas methods, not a
public Spark SQL API.