Corrected a race between the RecallTaskInjector and the TapeReadSingleThread

If the RecallTaskInjector had to do RAO query that takes too much time, the TapeReadSingleThread could start its infinite loop with no Recall task in it,
this would cause the stopping of the RecallTaskInjector. Thus, even if there are Retrieve jobs in the queue, as the RecallTaskInjector is stopped, the
TapeReadSingleThread will not be feed with Read tasks and will terminate. The remaining jobs in the queue will trigger a new mount