Update DNS-SD advertisements on fail-safe timeout (project-chip#23199)

When we remove a fabric during a pending add with fail-safe timeout, we
aren't updating DNS-SD adevertisements.  In addition to the obvious
consequence of continuing to advertise a now-stale identity, this can
also leave the advertisement slots full if the failed fabric addition
was to the Nth slot in an N-fabric device.

This then breaks advertisement when recommissioning to the Nth slot
thereafter for builds that do not set this:

CHIP_CONFIG_MINMDNS_DYNAMIC_OPERATIONAL_RESPONDER_LIST

For those builds, this condition persists until reboot or some other
action triggers DnssdServer::StartServer.

This commit fixes the problem by moving the DnssdServer::StartServer
call from the RemoveFabric invoke handler to the server's
OnFabricRemoved callback.  That will result in server restart both
for RemoveFabric invocations and fail-safe expiry.

Fixes project-chip#23169

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

@@ -272,11 +272,6 @@ const FabricInfo * RetrieveCurrentFabric(CommandHandler * aCommandHandler)
 CHIP_ERROR DeleteFabricFromTable(FabricIndex fabricIndex)
 {
     ReturnErrorOnFailure(Server::GetInstance().GetFabricTable().Delete(fabricIndex));
-
-    // We need to withdraw the advertisement for the now-removed fabric, so need
-    // to restart advertising altogether.
-    app::DnssdServer::Instance().StartServer();
-
     return CHIP_NO_ERROR;
 }
 
@@ -362,6 +357,10 @@ class OpCredsFabricTableDelegate : public chip::FabricTable::Delegate
     {
         ChipLogProgress(Zcl, "OpCreds: Fabric index 0x%x was removed", static_cast<unsigned>(fabricIndex));
 
+        // We need to withdraw the advertisement for the now-removed fabric, so need
+        // to restart advertising altogether.
+        app::DnssdServer::Instance().StartServer();
+
         EventManagement::GetInstance().FabricRemoved(fabricIndex);
 
         NotifyFabricTableChanged();