Add bounds fails with coordinates crossing the antimeridian or the Greenwich meridian.

[aulemahal]

Yet another bug with .cf.add_bounds... This one was already present before the 2D addition, but it would have been quite infrequent with 1D coordinates.

The interpolation / extrapolation estimates are not aware of the cyclic nature of the longitude.

MWE:

import xarray as xr
import cf_xarray

ds = xr.Dataset(
    {'data': (('x', 'y'), np.ones((3, 3)))},
    coords={
        'x': [1, 2, 3],
        'y': [1, 2, 3],
        'lon': (('x',), [170, -170, -150], {'units': 'degrees_east'}),
        'lat': (('y',), [-10, 0, 10], {'units': 'degrees_north'}),
    },
)

ds.cf.add_bounds(('lon', 'lat')).lon_bounds

yields:

array([[ 340.,    0.],
       [   0., -340.],
       [-160., -140.]])

but we would expect:

array([[ 160.,    0.],
       [   0., -160.],
       [-160., -140.]])

Sadly, a quick internet search hasn't given me any pointers towards tools that already take this into account, except for projection-based tools (like pyproj).

We could have a check for "longitude" that looks for jumps and wraps the coordinates, but that jump threshold would be arbitrary. It's nonetheless the simplest way out I see, before parsing grid_mapping into pyproj objects.