This is quite a simpel model, but ends up matching what we were able to do with
points-to.
I think this modeling excercise really shows that we need a bit of a different
way to model HTTP responses... but I'm not going to try to fix that in this PR.
The reason this becomes valueable right now, is that we can mark routed params
as taint-sources. Longer down the line, we can (hopefully) detect that a routed
param will only accept digits, and mark it safe for some of our taint-tracking
queries.