[opcreds] Fix LEAVE event on RemoveFabric (#18434)

* [opcreds] Fix LEAVE event on RemoveFabric

LEAVE event is not emitted on RemoveFabric or factory reset.
There are two reasons for that:
- Removing a fabric causes closing all active ReadHandlers
  for a given fabric, so we must flush the event log before
  that happens.
- There are two listeners for removing a fabric; the one
  that removes ACLs for a given fabric index is executed
  prior to the other one that generates LEAVE event. That
  causes "Accces control: denied" error.

Fix both issues.

* Add comment

src/app/clusters/operational-credentials-server/operational-credentials-server.cpp

@@ -364,8 +364,7 @@ class OpCredsFabricTableDelegate : public chip::FabricTable::Delegate
  static_cast<unsigned>(fabricIndex));
  fabricListChanged();
 
- // The Leave event SHOULD be emitted by a Node prior to permanently
- // leaving the Fabric.
+ // The Leave event SHOULD be emitted by a Node prior to permanently leaving the Fabric.
  for (auto endpoint : EnabledEndpointsWithServerCluster(Basic::Id))
  {
  // If Basic cluster is implemented on this endpoint
@@ -377,6 +376,26 @@ class OpCredsFabricTableDelegate : public chip::FabricTable::Delegate
  ChipLogError(Zcl, "OpCredsFabricTableDelegate: Failed to record Leave event");
  }
  }
+
+ // Try to send the queued events as soon as possible. If the just emitted leave event won't
+ // be sent this time, it will likely not be delivered at all for the following reasons:
+ // - removing the fabric expires all associated ReadHandlers, so all subscriptions to
+ // the leave event will be cancelled.
+ // - removing the fabric removes all associated access control entries, so generating
+ // subsequent reports containing the leave event will fail the access control check.
+ InteractionModelEngine::GetInstance()->GetReportingEngine().ScheduleUrgentEventDeliverySync();
+ EventManagement::GetInstance().FabricRemoved(fabricIndex);
+
+ // Remove access control entries in reverse order (it could be any order, but reverse order
+ // will cause less churn in persistent storage).
+ size_t aclCount = 0;
+ if (Access::GetAccessControl().GetEntryCount(fabricIndex, aclCount) == CHIP_NO_ERROR)
+ {
+ while (aclCount)
+ {
+ (void) Access::GetAccessControl().DeleteEntry(nullptr, fabricIndex, --aclCount);
+ }
+ }
  }
 
  // Gets called when a fabric is loaded into the FabricTable from storage

src/app/server/Server.h

@@ -333,21 +333,6 @@ class Server
  static_cast<unsigned>(fabricIndex), err.Format());
  }
  }
-
- {
- // Remove access control entries in reverse order. (It could be
- // any order, but reverse order will cause less churn in
- // persistent storage.)
- size_t count = 0;
- if (Access::GetAccessControl().GetEntryCount(fabricIndex, count) == CHIP_NO_ERROR)
- {
- while (count)
- {
- (void) Access::GetAccessControl().DeleteEntry(nullptr, fabricIndex, --count);
- }
- }
- }
- app::EventManagement::GetInstance().FabricRemoved(fabricIndex);
  };
 
  void OnFabricRetrievedFromStorage(FabricTable & fabricTable, FabricIndex fabricIndex) override