Bugs fixed:
 - There is a potential problem when mysqld waits for the ndb-handler
   components to be initialized. While waiting for connection
   to the ndb_mgmd we are not able to shutdown the mysqld, and might
   get stuck here forever if not able to make a ndb_mgmd connection.

 - Make the code more resilient by changing from infinite
   condition wait to a 1 second condition wait. This improves the
   chances that util thread and index stat thread will detect deinit
   and terminate themself properly.

 - The Ndb_cluster_connection_impl::connect_thread, which wait
   for the above ndb_mgmd connection, does *two* NdbSleep_SecSleep(1)'s
   instead if a singe as intended to.

Related cleanup:
 - Refactor the ndb-connection-wait into the new function
   ndbcluster_is_connected() and change it to use the (new)
   ndbcluster_cond instead of the Ndb_util_thread.COND.
   (ndbcluster_cond actually replace/rename COND_ndb_setup_complete)

-  The above refactoring allowed us to remove the global
   'extern' decl of ndb_util_thread, and make its COND
   and LOCK members 'private'

 - Refactor 'extern' decl. of variables related to above work
   by moving them from various *.cc files into a common *.h decl.