Allow backends to support find-and-lock in a single request

[betamatt]

This patch adds find-and-lock semantics to the active record backend and modifies the worker to use this method when the backend supports it. The idea is to allow workers to synchronize on the underlying database when the database supports it. As we've scaled out our worker count we've found contention between workers trying to lock jobs is a serious problem and generates a lot of extra query traffic in the current implementation. InnoDB's row level locks can instead be used to synchronize the worker polling so that each worker is guaranteed to get a lockable job, or none at all.

We haven't finished testing this yet but I'd like to start gathering feedback.

[bkeepers]

I actually just pushed a bunch of changes, one of which is adding a #reserve method. The worker calls this and expects it to find and lock an available job. It still does it in 2 queries though.

[bkeepers]

If you have time to rebase your changes and submit another pull request, it would be much appreciated.

[betamatt]

Brandon, I'll try to get to it early next week.

[betamatt]

Rebased: #167