improve xml / xpath testing

before changing the xml library (see #619), we should improve xml-related testing.
- some tests use xpath (lxml.etree based), but only in a "existence checking" way
- some tests just compare strings (xml fragments)

we should try to extract the interesting stuff using xpath and then compare structure/values somehow.

not sure how to do that precisely, suggestions welcome.