A corner case that wrongly casts to FloatIndex if the result is all-na (e.g. due to alignment) ``` import pandas as pd >>> idx = pd.Index(['a', 'b', 'c']) >>> idz = pd.Index(['d', 'e', 'f']) >>> idx.str.cat(idz, join='left') Float64Index([nan, nan, nan], dtype='float64') ```