Use geographiclib in lieu of pyproj

[ocefpaf]

Geographiclib is a stable, well maintained, and pure python library. We only use pyproj here for the great_distance calculation, so we should be OK with this change.

The results are similar and geographiclib algorithms do not suffer from this issue. See for a poor man's comparison https://nbviewer.org/gist/ocefpaf/7d290f94e5ffec1af560ded41d8c0dd1

Closes #72.

PS: only the doc build is failing. That should be fixed in #73.

[ocefpaf]

@kwilcox I'm pretty sure you have a thousand more efficient ways for doing this 😬. Feel free to change it to a way that suits ioos_qc better.

[kwilcox]

Vectorizing speeds it up a tiny bit. I played with numba but the optimization isn't good using Inverse in the loop and it ends up being slower.

def great_circle_distance_vectorize(lat_arr, lon_arr):
    def gc(y1, x1, y2, x2):
        return Geodesic.WGS84.Inverse(y1, x1, y2, x2)["s12"]
    dist = np.ma.zeros(lon_arr.size, dtype=np.float64)
    dv = np.vectorize(gc)
    dist[1:] = dv(lat_arr[:-1], lon_arr[:-1], lat_arr[1:], lon_arr[1:])
    return dist