Fix TopoJson non-collection geometries

[KhadeerBasha1232]

Fixes #1816

Problem

When a TopoJSON's objects entry had a type of Polygon or
MultiPolygon (instead of GeometryCollection), folium raised:

KeyError: 'geometries'

This happened because the code assumed every object had a geometries
key and didn't handle "bare" geometry objects that store their arcs
directly.

Fix

Updated the TopoJson handling in folium/features.py so it also
supports non-collection geometry types (Polygon, MultiPolygon,
etc.), not just GeometryCollection.

Testing

Reproduced the exact example from #1816 — it rendered without the
KeyError after this change. Ran the existing test suite locally with
python -m pytest tests --ignore=tests/selenium and all tests pass.

[KhadeerBasha1232]

This fixes #1816 — TopoJson raised a KeyError('geometries') when the
topology object's type was Polygon or MultiPolygon instead of
GeometryCollection, since the code assumed every object had a
geometries key.

Fix checks type first and falls back to treating the object itself
as a single-item geometry list when it's not a GeometryCollection.

Added a parametrized test covering both Polygon and MultiPolygon cases.
All existing tests + pre-commit checks pass locally. Happy to make any
changes if needed!

[adarshsm]

Thanks for picking this up, @KhadeerBasha1232 — the normalize-to-a-list approach is the right shape. A few things I hit while testing it against #1816:

1. Type-key robustness: geometry["type"] == "GeometryCollection" will itself raise KeyError: 'type' if an object omits type. Keying on the member you actually need — "geometries" in geometry — sidesteps that and reads as "does this hold sub-geometries, or is it one itself?"
2. Test could assert styling, not just rendering: since style_data's job is to attach properties.style, passing a style_function and asserting the style lands would catch a future regression that silently drops styling.
3. There's a second crash site this doesn't reach. style_data runs before children render, so once it's fixed, a single-geometry TopoJSON with a fields-based GeoJsonPopup/GeoJsonTooltip hits the same assumption again at features.py ~L1205 (["geometries"][0]["properties"]). Minimal repro:

   import folium
   from folium.features import GeoJsonPopup
   topo = {"type": "Topology",
           "objects": {"A": {"type": "Polygon", "arcs": [[0]], "properties": {"name": "A"}}},
           "arcs": [[[0, 0], [100, 0], [100, 100], [0, 100], [0, 0]]],
           "transform": {"scale": [1e-4, 1e-4], "translate": [9.47, 46.34]}}
   m = folium.Map()
   t = folium.TopoJson(topo, "objects.A")
   t.add_child(GeoJsonPopup(fields=["name"]))
   t.add_to(m)
   m.render()  # KeyError: 'geometries'

   I have a fix + regression tests for both sites ready — happy to push it as a follow-up PR once this lands, or you're welcome to fold the second-site fix in here. Whatever you and the maintainers prefer.

[KhadeerBasha1232]

Thanks for the thorough review, @adarshsm — really appreciate you digging into the edge cases and providing the repro script.

Pushed a follow-up commit (7157e9b) that addresses all three points:

Type-key robustness — switched the check to "geometries" in geometry instead of geometry["type"] == "GeometryCollection", so it no longer breaks on objects missing a type key. Added a regression test (test_topojson_non_collection_geometry_without_type) covering that exact case.
Second crash site — extracted a shared _get_geometries() helper and reused it in GeoJsonDetail, fixing the KeyError in GeoJsonPopup/GeoJsonTooltip for non-collection TopoJSON. Verified with your repro script, and added a parametrized test (test_topojson_non_collection_geometry_detail) covering both GeoJsonPopup and GeoJsonTooltip.
Styling assertion — updated the existing test to pass a real style_function and assert the style actually lands in properties.style, rather than just checking the map renders.

All local tests pass (pytest tests/test_features.py) along with pre-commit. Let me know if you'd like anything else adjusted — happy to iterate further.

[adarshsm]

That's a clean set of changes, @KhadeerBasha1232 — the shared _get_geometries() helper reads well, and the parametrized Popup/Tooltip tests cover the second site nicely. Looks complete to me, and pre-commit.ci is green. Thanks for the quick turnaround!