> Regarding the use of PY_SSIZE_T_CLEAN, I assume that Py_ssize_t is to be
> preferred over plain ssize_t. Is this correct?

Yes, ssize_t doesn't exist everywhere AFAIK.
(size_t does, or at least we assume it does)

> Also, I was wondering whether I need to add some sort of license boilerplate to
> the beginning of bz2.py? With _bz2module.c, I presume I should retain the
> copyright information from the old bz2module.c. Would something like this be ok?

Well, I would personally advocate not re-adding a license boilerplate,
since it doesn't serve a purpose (nearly all of Python is freely usable
under the PSF License, and the authors are documented by version
control).
You could ask on python-dev to get other opinions, though.