RotatedPole declared with ellispoid, but projection still is spherical

[aulemahal]

Stemming from findings I made after the change of PROJ 9.8 (see issue #2634), I realized that the default Globe for RotatedPole is incorrect, or at least misleading. See discussion in OSGeo/PROJ#4714.

In crs.py, if no globe argument is passed to the RotatedPole constructor, it defaults to this:

cartopy/lib/cartopy/crs.py

Line 1986 in 05e5c20

globe = globe or Globe(semimajor_axis=WGS84_SEMIMAJOR_AXIS)

And the default globe uses ellipse='WGS84'. However, as confirmed by a maintainer of PROJ the ob_tran projection is always spherical, no matter what +ellps arg is given. Thus, we have a potentially confusing definition.

If possible, I would suggest that the default globe be spherical. For example, the line above could be changed to:

globe = globe or Globe(ellipse='sphere', semimajor_axis=WGS84_SEMIMAJOR_AXIS)

[greglucas]

@dopplershift it might be worth thinking about RotatedPole in your work on #2653 and what the expectations should be. I'm not sure I understand all of these subtleties myself and what should happen in the various cases for defaults.