WL#2703 restart for ndb_mgmd

Solving two problems with one stone.

Allows the stopping and restarting of mgm nodes other than the one the mgmclient
is connected to.


ndb/include/mgmapi/mgmapi.h:
  Add the internal ndb_mgm_end_session command.
  
  This will unreserve the nodeid we have allocated synchronously.
  
  Otherwise we can't do a restart of a node really quickly as the nodeids are cleaned
  up after the connection to mgmd is closed.
ndb/include/mgmcommon/ConfigRetriever.hpp:
  Allow configuration on if end_session is going to be called on object destruction.
  
  We need to set this to false for ndbd as we fork()
ndb/src/common/mgmcommon/ConfigRetriever.cpp:
  When destroying ConfigRetreiver, ndb_mgm_end_session - i.e. deallocate the nodeid
ndb/src/common/util/SocketServer.cpp:
  When destroying a SocketServer, close the server socket.
ndb/src/kernel/main.cpp:
  don't purge allocated resources when cleaning up in parent process (nodeid)
ndb/src/kernel/vm/Configuration.cpp:
  have option end_session to closeConfiguration
ndb/src/kernel/vm/Configuration.hpp:
  have option end_session to closeConfiguration
ndb/src/mgmapi/mgmapi.cpp:
  Implement ndb_mgm_end_session
ndb/src/mgmclient/CommandInterpreter.cpp:
  Correct output of STOP as we can now stop mgmd as well as ndbd
ndb/src/mgmsrv/MgmtSrvr.cpp:
  Add code into start for connecting to our own mgmd.
  
  Create sendStopMgmd() which does the same job as sendSTOP_REQ, but for ndb_mgmd
  
  Allow stopping of other ndb_mgmd processes by creating a connection to them
  and issuing the stop command
  
  When stopping all nodes, stop other ndb_mgmd processes as well.
  
  Remove set_connect_string. Replace with connect_to_self. This is a much better
  way of doing things.
ndb/src/mgmsrv/MgmtSrvr.hpp:
  add connect_to_self and remove set_connect_string.
ndb/src/mgmsrv/Services.cpp:
  Add endSession.
  
  - delete Allocated_resources for this connection
  - create new Allocated_resources for this connection
  
  conceivably you could keep the socket open across node restarts (and even
  possibly get a different node id). But I wouldn't try it and expect happiness.
ndb/src/mgmsrv/Services.hpp:
  Add endSession
ndb/src/mgmsrv/main.cpp:
  allow mgmd to be restarted.
  
  - add g_RestartServer flag
  - move connecting to our own mgmd into MgmtSrvr (where it belongs)
  - output correct Shutdown/Restart message on shutdown/restart