In general on Python 2, simplejson decodes ASCII strings as str, only promoting
to unicode when needed:

>>> simplejson.loads('["Spaetzle", "Spätzle"]')
['Spaetzle', u'Sp\xe4tzle']

Since 83a493db6a8b859ec7b10fa85365dd3fdf144c68, though, simplejson has always
decoded empty JSON strings as unicode:

>>> simplejson.loads('""')
u''

This PR restores the old behavior of decoding empty strings as str.

* Define less version-specific macros.
* Remove redundant version checks and code inside version-specific code.
* Always check `PyUnicode_READY()` for external unicode objects.
* Use `PyUnicode_Join()`.
* Use `PyUnicode_FromOrdinal()`.
* All Python string constants are now allocated only once.