Here is a patch for the commit fest.
I have left the pg_upgrade script and just put in the new call instead
of the three old ones.
For the 8.3 case, which calls for a real vacuum in addition to the
analyzes, I have simply added another vacuumdb call to the script. This
technically makes the script do more work than before, where the vacuum
was part of the final analyze, but I don't see an elegant way to do it
otherwise.