Pandas first_valid_index for all NaN DataFrame

Similar to this one, but on a non-empty DataFrame whose values are all NaN.

This raises an error, instead of returning None

pd.Series([np.nan]).first_valid_index() # this returns None
pd.DataFrame([np.nan]).first_valid_index() # raise error

>>> pd.__version__
u'0.18.1'

Comment From: chris-b1

Thanks - added the case to #9752, same underlying issue.