Fix NRE in RavenDb durability recovery

[Bishbulb]

Summary

I ran into this running multiple Wolverine services against a single shared RavenDb envelope database. Every durability recovery pass was throwing a NullRef exception and spamming the logs every few seconds. I traced it to RavenDbDurabilityAgent.tryRecoverIncomingMessages.

Root cause

The method scans all distinct ReceivedAt URIs in the IncomingMessage collection and calls FindListenerCircuit for each. When a URI belongs to a queue that this node doesn't service (because a different service owns it), FindListenerCircuit returns null which causes an NRE trying to access circuit.Status:

  var circuit = _runtime.Endpoints.FindListenerCircuit(receivedAt)!;
  if (circuit.Status != ListeningStatus.Accepting)

Note that I believe the same issue would occur in a single-service deployment if a listener URI was removed from config (or renamed), at that point any inbox documents that were still referencing the old URI would result in the same null reference exception.

Fix

Null-check the circuit variable and skip it

[jeremydmiller]

Thanks for the clean fix and regression test.