Regression in addcyclic behaviour

Pull request #138 changed the behaviour of basemap.addcyclic. The data and longitude arrays are now being treated the same, i.e. the first point being appended to the end, but previously the longitude array was handled by adding the difference between longitudes to the last longitude, which is in my mind the correct behaviour.

An example:

import numpy as np
from mpl_toolkits.basemap import addcyclic

lons = np.arange(0, 360, 60)
data = np.ones([5, len(lons)]) * np.arange(len(lons))
new_data, new_lons = addcyclic(data, lons)
print new_lons

Prior to #138 being merged this would result in [0, 60, 120, 180, 240, 300, 360] but after it results in [0, 60, 120, 180, 240, 300, 0].

Start by reading the implementation of basemap.addcyclic and the change introduced by pull request #138. Reproduce the issue with the NumPy example in the report, then verify that data receives the cyclic point while the longitude array ends at 360 rather than repeating 0.