GeoPandas: Preserve CRS metadata for empty geometry results

[jiayuasu]

Expected behavior

Distributed GeoSeries and GeoDataFrame objects should preserve explicitly
assigned CRS metadata when their geometry rows are empty, all null, or carry
SRID 0.

Explicit set_crs values should take precedence over inherited lineage
metadata. Copies, GeoPandas conversion, and construction from an existing
geometry column should retain that metadata, while switching to a different
active geometry column must not leak the previous column's CRS.

Actual behavior

Sedona normally derives CRS from the SRID stored in geometry rows. Empty and
all-null results have no non-null geometry from which to recover that SRID, so
CRS can be lost after construction, copy, conversion, or a distributed
operation that removes every row.

The same fallback mechanism is currently being introduced independently by
#3183, #3184, and #3185. Keeping it in those feature PRs duplicates shared
state-management code and creates avoidable merge conflicts.

Steps to reproduce the problem

from sedona.spark.geopandas import GeoDataFrame, GeoSeries
import pyspark.pandas as ps

series = GeoSeries([None], name="shape", crs="EPSG:4326")
with ps.option_context("compute.ops_on_diff_frames", True):
    frame = GeoDataFrame({"value": [1]}).set_geometry(series)

assert series.crs == "EPSG:4326"
assert frame.crs == "EPSG:4326"
assert frame.copy().crs == "EPSG:4326"
assert frame.to_geopandas().crs == "EPSG:4326"

The shared implementation should also cover zero-row derived results,
duplicate-index construction, explicit CRS removal or replacement, deep
GeoSeries copies, and active-geometry switches.

Settings

Sedona version = 1.9.1-SNAPSHOT

Apache Spark version = 3.4+

API type = Python GeoPandas API

Python version = 3.8+

Environment = Standalone