Series.cummin raises ValueError when datetime64 Series contains NaT

Ref: http://stackoverflow.com/q/27240365/190597

```
s = pd.Series(pd.date_range('2008-09-15', periods=10, freq='m'))
s.loc[10] = pd.NaT
s.cummin()

ValueError: Could not convert object to NumPy datetime
```
- validate for `cummax` & `timedelta64[ns]`
