What I've done here is to instrument sup-sync to print the text
and terms objects it constructs just before indexing a message.
Then I've made my g_mime_test program achieve (nearly) identical
output for an example email message, (just missing the body
text). Next we can start shoving this data into a Xapian index.
