This was breaking integration tests but affects all zero-latency server situations. There's a good description in code, but the basics are a race between a broadcast (which goes first) and the post-SLAVEOF reconfig (which loses) creates a situation where the proper topology isn't seen until the next reconfig time span pass.

We now prevent that broadcast (we're broadcasting to ourselves) from running on ourselves because we're about to run another one anyway. This doesn't 100% eliminate the race because there's a minute chance of a pub/sub landing between the 2 if blocks here in an external Interlock. But the chance is crazy small now, at least.

Note: the logging of *why* a reconfig didn't run is now enhanced as well. We log what was trying to run and what was blocking it in the output for much easier debugging next time.