caused by a self._input.readline() call that wasn't checking for the
NeedsMoreData marker.

msg_43.txt contains a message that illustrates the problem, when
email.message_from_*() is called.  That interface uses the Parser API, which
splits reads into 8192 byte chunks.  It so happens that for the test message,
the 8192 chunk falls inside a message/delivery-status, which is where in the
FeedParser the readline() call was that didn't check for NeedsMoreData.

I also added an assert to unreadline() so it'll be more evident if an attempt
to push back NeedsMoreData ever happens again.

Bump the email package version number.

(First draft of patch contributed by Steven Bethard.)

(Submitted by some anonymous person with an amazing eye for grammer nits.)

Will backport.

Python was installed with the 'only for me' option. The registry key
had a hardcoded '2.3' in it where the python version chosen for
installation should be used instead.

Will backport myself.

  the beginning of a paragraph do not generate errors; this affected
  things like \refmodule when it came first in a paragraph
- clean up the .sty file to separate out the treatment of the start
  of a new paragraph

- don't create .tgz version of doc source package, since we're not
  generating that for the other formats either

so the docs can be built.