Series.str should only be defined for strings, not all Series with dtype=object

[shoyer]

#9322 will make Series.str raise an exception if it is accessed on Series instances with non-object dtype. In principle, the exception should really be raised for any non-strictly string-like data, but that's not practical until pandas has a true string dtype to use (currently we abuse np.object_ for this purpose).

[jreback]

closing in favor of #13877