bug: Sedona ST_Crosses doesn't support GeometryCollection

```sql
select st_crosses(st_geomfromtext('linestring (1 1, 1 2)'), st_geomfromtext('geometrycollection (point (1 1))'));

Caused by: java.lang.IllegalArgumentException: Operation does not support GeometryCollection arguments
at org.locationtech.jts.geom.Geometry.checkNotGeometryCollection(Geometry.java:1735)
at org.locationtech.jts.geom.GeometryRelate.relate(GeometryRelate.java:203)
at org.locationtech.jts.geom.Geometry.relate(Geometry.java:1006)
at org.locationtech.jts.geom.Geometry.crosses(Geometry.java:803)
at org.apache.sedona.common.Predicates.crosses(Predicates.java:47)
at org.apache.spark.sql.sedona_sql.expressions.ST_Crosses.evalGeom(Predicates.scala:176)
at org.apache.spark.sql.sedona_sql.expressions.ST_Predicate.eval(Predicates.scala:61)
```

Once this is fixed, we should remove [this workaround](https://github.com/apache/sedona/blob/9ea7c23633cf97b30a662465aa061feecfdaa79c/python/sedona/spark/geopandas/geoseries.py#L1221-L1230) in geopandas.

Start with org.apache.sedona.common.Predicates.crosses and the Scala ST_Crosses path shown in the stack trace, then inspect the GeometryCollection workaround in python/sedona/spark/geopandas/geoseries.py at lines 1221-1230. Verify the SQL example no longer raises for GeometryCollection arguments, and remove the workaround once the predicate supports it.

Written by the indexing model from the issue text.