Fix three stale documented defaults (TopoJson.overlay, TimeSliderChoropleth.overlay, GeoJsonPopup.localize)

[dylanpulver]

Three classes override a default and kept the parameter description they were copied from, so the documentation states the old value:

	documented	actually defaults to
TopoJson.overlay	False	True
TimeSliderChoropleth.overlay	False	True
GeoJsonPopup.localize	False	True

In each case the code is right and the docstring is stale, and the tie is broken by a sibling in the same file rather than by preference:

• Layer (map.py) both documents and defaults overlay=False. GeoJson overrides it to True and documents True. TopoJson and TimeSliderChoropleth override it to True and kept Layer's line.
• GeoJsonTooltip documents localize as False and defaults to False. GeoJsonPopup overrides to True and kept the tooltip's line.

overlay decides whether a layer shows up in LayerControl as a toggleable overlay or as a base layer, so this is a wrong mental model rather than a cosmetic slip.

The new test walks every class in the package — 96 of them — and compares the documented boolean default with the signature default for overlay, control, show and localize. It is restricted to those four because a bare True/False in a docstring is unambiguous to compare, whereas the numeric and string defaults elsewhere are written in too many styles to check mechanically without false positives.

Verification: with the three docstrings reverted to their current text, exactly those three parametrised cases fail and the other 94 pass, so the check is neither vacuous nor over-broad. There is also a small self-check asserting the docstring parser really does detect a mismatch. black 26.5.1 and ruff 0.16.3 (the versions pinned in .pre-commit-config.yaml) report the three files unchanged and clean.

Suite run locally, Python 3.14, excluding tests/selenium: 369 passed, 2 failed — test_icon_invalid_marker_colors and test_valid_png_size, both of which fail identically on unmodified main here, so they look like environment rather than regressions. tests/snapshots could not be collected at all locally because pixelmatch is not installed; that path is untested by me.