Use platform nodes + route membership for bus stop clustering

Replaces distance-based clustering with route-membership grouping:
- download_osm.py: collect route_refs per stop node; use platform roles
  for bus stops, stop roles for rail stops
- prepare.py: carry route_refs through to master.gpkg
- render.py: group_by_route_membership() uses union-find to cluster
  same-name stops sharing at least one route into one marker; stops on
  disjoint routes are kept separate. No distance threshold needed.

Results: 786 bus platforms -> 465 markers (was 1610 -> 781).
Veksøvej: 2 platforms sharing routes 2A+5C -> 1 marker.
Peter Bangs Vej: 7 platforms on routes 10/21/22/4A/7A -> 2 markers.

copenhagen/scripts/prepare.py

@@ -171,7 +171,7 @@ def main():
     if stops is not None and not stops.empty:
         stops = stops[~stops.geometry.isna() & stops.geometry.is_valid]
         stops_clipped = gpd.clip(stops, area_geom)
-        keep = ["mode", "style", "name", "ref", "public_transport", "railway", "geometry"]
+        keep = ["mode", "style", "name", "ref", "public_transport", "railway", "route_refs", "geometry"]
         for c in keep:
             if c not in stops_clipped.columns:
                 stops_clipped[c] = None