DOC: describe geometry_area_perimeter size limit concretely

[Htet-Kaung-San]

Closes #1592.

The note on Geod.geometry_area_perimeter said only:

There are a few limitations :

• only works with areas up to half the size of the globe ;
• certain large polygons may return negative values.

As #1592 points out, that doesn't say what "large" or "certain" mean — whether it's city-sized or continent-sized, and whether shape, poles or the antimeridian are involved.

Both bullets turn out to be the same rule. The area is reduced modulo the total area of the ellipsoid into (-A/2, A/2], so a region covering more than half the ellipsoid is reported as the negated area of its complement:

from pyproj import Geod
from shapely.geometry import Polygon

geod = Geod(ellps="WGS84")
nearly_global = Polygon([(-179, -89), (179, -89), (179, 89), (-179, 89)])
print(geod.geometry_area_perimeter(nearly_global)[0])
# -2833262901265.9062

That polygon is wound counter-clockwise and encloses 99.4% of the globe, but comes back as roughly -0.56% — the thin strip left outside it, negated.

So the answers to the two questions in the issue are concrete:

• "large" means precisely "encloses more than half the ellipsoid" — about 2.55e14 m² on WGS84.
• "certain" isn't about shape, proximity to a pole, or crossing the antimeridian. Size alone decides it.

I checked the (-A/2, A/2] bound holds across a grid of boxes spanning ±10° to ±179.9° longitude and ±10° to ±89.9° latitude.

The docstring already carries a separate warning that traversal direction is signed, CCW positive. That's the other reason a result can be negative, and unlike the size limit it applies at any size, so I cross-referenced the two rather than describing them independently.

Changes

• Replace the vague note with the modulo rule, the WGS84 numbers, and an explicit statement of what does not matter.
• Cross-reference the traversal-direction warning as the other, size-independent cause.
• Add doctests for the wraparound and for recovering the enclosed area by adding the total ellipsoid area back.
• Changelog entry.

Documentation only — no behavior change.

Testing

flake8, ruff and codespell are clean on the changed files, and blacken-docs --skip-errors leaves them unchanged. The docstring's doctests pass (14 examples, 0 failures), including the two new ones. The 99.4% figure is 99.4445% before rounding, so it isn't sitting on a rounding boundary.