support DelayedVector in filter_spatial

[soxofaan]

Using remote URL in filter_spatial came up in https://discuss.eodc.eu/t/http-413-error-on-increasing-data-size-in-demo/286/11
but we don't support that yet:

openeo-python-driver/openeo_driver/ProcessGraphDeserializer.py

Lines 1068 to 1074 in 75c634f

	def filter_spatial(args: Dict, env: EvalEnv) -> DriverDataCube:
	cube = extract_arg(args, 'data')
	geometries = extract_arg(args, 'geometries')
	if not isinstance(geometries, dict):
	# TODO: support DelayedVector
	raise NotImplementedError("filter_spatial only supports dict but got {g!r}".format(g=geometries))

[m-mohr]

Probably because it's not defined as such in the process: https://processes.openeo.org/#filter_spatial

[soxofaan]

DelayedVector is an implementation detail that would allow to use a geojson-compatible geometries argument that is lazy loaded from a file or URL (e.g. from our custom read_vector process, much like the proposed new load_files process).
We already support that in aggregate_spatial and could do that in filter_spatial too