This is to fix SEDA exports tests on python3.

We first dump XML documents as unicode using the _encoding argument
introduced in previous changeset. Accordingly, expected files are
changed to remove the XML declaration because dumped values will not
contain them.

Then we open files in text mode (using io.open() for consistency between
python2 and python3) and do string formatting using unicode values.
(Especially, six.binary_type(<some integer>) produces different values
in python2 and python3, and we actually want to stringify the integer --
the eid value.)