BUG: Series _transform_fast fails for datetime with null groups

[chris-b1]

xref #10972 - but not the same issue. I've got a fix for this + #12737 coming.

Code Sample, a copy-pastable example if possible

In [20]: df = pd.DataFrame({'grouping':[np.nan,1,1,3], 
                            'v':[1.1, 2.1, 3.1, 4.5], 
                            'd':pd.date_range('2014-1-1','2014-1-4')})

In [21]: df.groupby('grouping')['d'].transform('first')
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-21-cb1ed73aabc3> in <module>()
----> 1 df.groupby('grouping')['d'].transform('first')

C:\Users\Chris\Anaconda\lib\site-packages\pandas\core\groupby.pyc in transform(self, func, *args, **kwargs)
   2738                 # cythonized aggregation and merge
   2739                 return self._transform_fast(
-> 2740                     lambda: getattr(self, func)(*args, **kwargs))
   2741
   2742         # reg transform

C:\Users\Chris\Anaconda\lib\site-packages\pandas\core\groupby.pyc in _transform_fast(self, func)
   2781         out = func().values[ids]
   2782         if not mask.all():
-> 2783             out = np.where(mask, out, np.nan)
   2784
   2785         obs = np.zeros(ngroup, dtype='bool')

TypeError: invalid type promotion