distinguish known-bad targets from random failures

[Habbie]

In our job matrix (currently 92 big) there are a few targets that really fail at build-time for a few reasons (that we should enumerate). Let's call those "known bad targets" for our specific purpose. Then, sometimes specific target/version combinations fail because, say, git.openwrt.org was having a moment. We currently have no way to distinguish these two failure modes from each other other than manually going over each one.

We should, once, and then periodically, go over all, and decide which targets we can do without (because they fail consistently), and fail the build if any other targets didn't pass the build for transient reasons.