-Fix problem where Ndb_cluster_connection::connect always
  returned -1 when it failed to allocate nodeid. Now returns 1 if
  the error is not fatal.

 - Fix a problem where Ndb_cluster_connectin::connect had hardcoded
  values for "no_retries"  and "retry_delay_in_seconds". Now those two
   values how long the function will retry both connect and alloc nodeid

 - Fix bug in ConfigRetriever::allocNodeId, where the return code
  of ndb_mgm_connect was interpreted wrong, causing the function
  to return or retry even if the connect suceeded. Refactor function slightly.

 -  Add testcase for Ndb_cluster_connection::connect() returning 1
   when it can't connect to ndb_mgmd and also when it get a non fatal error
   returned while allocating nodeid.