MongoDB replication should query the oplog in a separate thread

Blocking the processTimer method of the computation may lead to unneeded rebalance on the Kafka side. Moreover, by blocking the method, we are also not able to askCheckPoint regularly and thus increases the RPO to unmanaged values.

In order to avoid that, the tailable cursor on the oplog.rs collection must be run in a separate thread. The processTimer will then continuously taking the found oplogs and send them to the stream.