Communicate via durable store?

[adammck]

We currently communicate between rangerd and rangelets via gRPC. But we write the range and placement states to durable storage (via the Persister; only Consul, for now) first, so the RPCs could simply be considered a notification that the contents of the store have changed. As such, the rangelets could also just watch the store for the changes, and react accordingly. Would need to index placements by nodeID rather than inside of ranges, to avoid nodes having to read the entire range map.

Unsure how the nodes would communicate their actual state (RemoteState) and health back to rangerd in this scenario. Maybe still RPC probes?