[exim] Testing an interesting spam catching theory

I have a theory that if a person ran a bayesian filter only on the
headers that it would be more accurate that running it on the body of
the message as well. And - I'm in the process of trying to test it using
spamprobe but feeding spamprobe only headers. I'm using Spam Assassin to
select the headers that it will auto train on.

If it works - I will share with all.