Shiftdata breaks pcolormesh wrapping

Nobody has claimed this yet.

I have a data grid which spans the whole globe from -180 -> 180. pcolormesh can plot this OK if I have lon_0=0 in which case Basemap doesn't have to do any wrapping

However if I do this for lon_0=180 (where Basemap has to shift some data), I get the following incorrect plot:

The code I have used for this is (using 1.0.7):

```
import numpy as np
from mpl_toolkits.basemap import Basemap
import matplotlib.pyplot as plt
x = np.arange(-180, 190, 10)
y = np.array([40, 50, 60])
data = np.arange(72).reshape((2, 36))
xv, yv = np.meshgrid(x, y)
# Add an empty row and column of data so that the data is the same shape as X,Y
wider_data = np.ma.zeros(xv.shape)
for index, v in np.ndenumerate(data):
wider_data[index] = data[index]
for lon in (0, 180):
map = Basemap(lon_0=lon)
map.drawparallels(np.arange(-90, 90, 20), labels=[1, 1, 0, 1])
map.drawmeridians(np.arange(-180, 180, 30), labels=[1, 1, 0, 1])
map.pcolormesh(xv, yv, wider_data, latlon=True)
map.drawcoastlines()
plt.show()
```

I suspect that this problem is related to https://github.com/matplotlib/basemap/issues/182 because to get around that issue, I have had to add an empty row and column to the data. This should still be OK though, since that extra data should just be ignored (from the pcolormesh docs).

However, what actually happens is that the empty row gets moved around because shiftdata doesn't know it should ignore and so the plot is wrong.

Start by running the provided Basemap 1.0.7 reproduction with pcolormesh and compare the lon_0=0 and lon_0=180 plots. Inspect the shiftdata path used by pcolormesh wrapping; done means the extra row and column remain ignored and the shifted plot matches the expected global grid.

Written by the indexing model from the issue text.