this was kinda completely broken, oops. When we first hit an
image with a score > 0 we put its *URL* in the univs dict,
then for each later image for the same arch we'd compare its
*score* against the current candidate's *URL*, which was...not
what I meant to do. strings always win vs. ints, so the effect
of this was that we'd always run the universal tests on the
first image we found for the arch with a score > 0.
We also lied about what image we were running the universal
tests on, because when we tried to log the URL we used the
var url, but we didn't read that var from the univs dict;
we were actually using the value of url that leaked from the
previous loop over the images list, image list, which was just
the URL of whatever the last image found was.
Oops!