Support raster cube input to run_udf via STAC catalog materialization

[Copilot]

When a save_result node feeds into run_udf, the datacube needs to be materialized to disk and passed as a STAC catalog URL. Previously, run_udf couldn't accept a DriverDataCube as input.

Changes

• GeopysparkDataCube now implements SupportsRunUdf — the ProcessGraphDeserializer.run_udf() detects this interface and dispatches directly to the cube's run_udf method
• run_udf() materializes the cube to a STAC catalog — calls write_assets() to produce GeoTIFFs, then writes a STAC 1.0.0 catalog linking the items, and passes the catalog URL as structured data to the UDF
• _write_stac_catalog() helper — writes valid STAC catalog + item JSON files from write_assets output

UDF receives the catalog URL like this:

from openeo.udf import UdfData, StructuredData

def udf_function(data: UdfData):
    stac_url = data.get_structured_data_list()[0].data["stac_catalog_url"]
    # Process STAC catalog (e.g., invoke CWL workflow)
    return UdfData(structured_data_list=[StructuredData(data=result, type="dict")])