Reactions now put themselves in a queue when they get a callback from an atom. The queue is flushed before rendering, and on deref on individual reactions. This turns out to perform much better in many cases.
And use in test for with-let