The poller currently scans the db, looking for situations that it can try to recover from. We should add one where it looks for modules stuck in the wait state.

We got there today, when @ignatenkobrain submitted a build of testmodule (module build 1022). The queue was backed up, we restarted it, and the signal from the frontend to the backend was lost. Adding this to the poller is the simplest way to resolve this in the future.