Recently, there have been multiple instances where we turned down builders. As part of this we either (a) removed the *-pre-cq, or (b) *-pre-cq just broke due to bitrot.

In either case, a bunch of CLs that contained that -pre-cq config in its target set got stuck until user action or verification timeout (weeks, iirc).


Instead, we should:
- Recompute target set each time we're about to launch pre-cq jobs / validate a CL
- Find what needs to be done (we don't need to reset the passed set)
- Launch / validate against the remaining.


pre-cq is a class-A cros-infra service, and this bug is causing a non-trivial number of CLs to get stuck until developer action ( I've heard of people complaining about rebasing not fixing this problem, but I don't believe them ;) )
Also, I'd expect this problem to only grow as our board churn isn't going to reduce any time soon.

So, Pri-1.