fix behaviour of ALL STOP and SHUTDOWN in relation to MGM nodes


ndb/src/mgmclient/CommandInterpreter.cpp:
  Shutdown management servers properly (one we're connected to last).
  
  Fix potential problems if disconnect from one mgmd and connect to another (that's a different cluster)
ndb/src/mgmsrv/MgmtSrvr.cpp:
  fix regression in how nodes are stopped
  ALL STOP - db nodes only
  SHUTDOWN - db and mgm nodes
  
  fix race where mgmd could stop itself before it sends the response to stop.
  In ~MgmApiSession() we now do the disconnect.
  
  We have (in various places) a return stopSelf which tells the caller if they
  need to stop or restart this daemon.
ndb/src/mgmsrv/MgmtSrvr.hpp:
  add stopSelf return parameter to stopNodes and restartNodes.
  
  Rename stop to shutdownDB as this name better reflects what it does
  
  Rename restart to restartDB as this name better reflects what it does
ndb/src/mgmsrv/Services.cpp:
  Stop or restart server only on mgm session disconnect (of session that asked us to)
ndb/src/mgmsrv/Services.hpp:
  add m_stopSelf member for tracking what we should do