clogd: Fix bug 474179 (and still investigating others)
    
    Other bugs may include: 471042
           and less likely: 464550
    
    3 additions are in this checkin:
    - Fixed a case where a node exiting the CPG would not
      by able to send a 'checkpoint ready' notification,
      because the attempt is made after leaving the CPG.
      This can leave an incoming node with now way of getting
      the log information necessary to get going.
    
    - If a node quickly enters, exits, and enters a CPG.  It
      is possible for other nodes in the cluster to fill a
      request for checkpoint data for the first 'enter', but
      *after* the node has left.  This information would then
      collide with new checkpoints that should happen on the
      next 'enter'.  We fix this by clearing any checkpoint
      data just before joining the CPG on entry.
    
    - Add more debugging information that is printed out when
      'clogd' is signaled with USR1.