- This pushes the units conversion up to PROJ and allows a globe
  to be defined with a WGS84 semimajor axis which helps for
  coordinate conversions down the line.

- Set trap=False in transform_points to allow out of bounds calculations
- Ignore the geocentric projection pivot in the image transform

Update the tests to use the proper longitude/latitude formatters
rather than rolling our own.

Moves the transform_points update from PlateCarree into the
Projection superclass and tests for _CylindricalProjection in
there. Also added RotatedPole to the trace.pyx mod-180 checks.

You can't draw coastlines associated with WGS84 on a smaller ellipse
in newer PROJ versions due to different celestial bodies error.

…ntric CRS for more accurate datum information

Some projections can be geographic in pyproj, but Cartopy views
GeodeticCRS different and has some different handling of functions
based on this check. For now, we will force all Projections to not
be of Geodetic type.

The longitude and latitude were in the wrong locations, so put
these into the right X/Y spot for the test.

The old array is still used for determining wraps later on, so we
don't want to modify the array in-place.

Rather than updating the image, just bump the tolerance up a tiny amount
to account for the shift in some of the pixels. This will still
catch the big errors like missing a wrap.

When setting the extent, you have to be within the proper domain,
so subtract 360 to force the proper bounds.

PROJ requires semimajor axes to be compatible, but Cartopy previously
used an artificial semimajor axis to make the units work out nicely.
Instead make the units work out by using the "+to_meter" argument
to PROJ.

To split the Polygon we need to go outside of the True bounds,
not just land on it, so increase the limit just a bit.

Also update the expected values in the test for gridding data
outside of the transform. Once the values are wrapped, the
expected value should be the 3rd entry.

Updated the tolerances of a few, but the ones that changed by
a decent amount were updated.

Globe is not the first argument, so it needs to be passed as a keyword.

- Updating the extents of the image to correspond with the new
  PlateCarree "to_meter" conversion to always be between -180, 180.
- Swap the extents in y to be negative so that the image is the
  same orientation as in the URL.
- Remove the Frame from the Axes as the boundary isn't computed
  properly for an eccentric Globe.

PROJ shortcuts the transform if the datum is the same, so remove the
datum to get a spherical transform in this example.

These are in accordance with NEP29 for a release after July 2021.

- Removes old images no longer supported for testing.
- Updates tolerances to support older versions of MPL on those tests.

Adjust the image tolerances for the minimum MPL version. The images
all look roughly the same, only with minor pixel shifts.
This reduces the number of images added to the repository.

- Removed some duplicate transformations
- Cleaned up documentation

This reverts commit 3b39029.

Adds the environment variable PYPROJ_GLOBAL_CONTEXT=ON

Co-authored-by: Ryan May <rmay31@gmail.com>

It is slow to create an interpolator and there are a lot of calls to
project, so keep the interpolator around to speed up repeated calls.

pyproj == comparison is slow, so if we are comparing two Cartopy
CRSs, return a fast-path comparison. Let the pyproj comparison
handle the other cases.