Correlation() gives up too easily

[CommonClimate]

It's a tough world out there, and sometimes we need to compute hundreds of correlations at once.
We can't have Series.correlation() error out with a whimper if it can't find any overlap between Series X and Y. It needs to put on its big boy pants, return NaNs for r, p and signif=None, so that if we run MultipleSeries over a large MultipleGeoSeries object, one faulty dyad won't doom the whole thing. I'll allow it to whine a warning, though.

Context: currently using Pyleoclim to filter records in a large database of > 650 records for PReSto.