Add "count distinct" function, useful for groupby

[wesm]

something like

def dist(x):
       return len(np.unique(x))

[changhiskhan]

len(np.unique(self.dropna())) ?

with a Cython distinct_count_level_1d function?

[wesm]

Maybe this should just be a groupby instance methods-- like Series now has the value_counts method so you can do:

len(s.value_counts())

but in a groupby scenario, you could do:

s.groupby(grouper).agg(lambda x: len(x.value_counts()))

but that's a bit verbose. Maybe like

s.groupby(grouper).nunique()