Ensure that READY checking behavior doesn't result on constant rebalancing

[tstirrat15]

See the commit message here: authzed/spicedb@94145c0

The hashring was originally written the way that we changed it to in #4, and we moved off of that implementation because it was causing too much thrash, as connections went from READY to not READY relatively frequently.

We may need to rethink the balancing approach.

[tstirrat15]

It also may not be an issue if the hashring remains stable and the current behavior just routes past a stuck subconn to the next connection in the ring.