This splits the `cpp/overruning-write` into two separate queries based
off on the reason for the estimation. If the overrun is detected based
on non-trivial range analysis, the results are now marked by the new
`cpp/very-likely-overruning-write` high precision query. If it is based
on less precise, usually type based bounds, then it will still be marked
by `cpp/overruning-write` which remains at medium precision.
