-This is mostly fixes for correct behaviour when using query cache + transactions + the thread that
fetches commit count from NDB at regular intervals. The major fix is to add a
list in thd_ndb, that keeps a list of NDB_SHARE's that were modified by
transaction and then "clearing" them in ndbcluster_commit.


mysql-test/r/ndb_cache2.result:
  Updated test cases for the ndb_util thread, more simultaneous tables and more tesst
mysql-test/t/ndb_cache2.test:
  Updated test cases for the ndb_util thread, more simultaneous tables and more advanced tesst
sql/ha_ndbcluster.cc:
  Add table changed during transaction to list of changed tables in Thd_ndb, this list is then used in ndbcluster_commit to invalidate the cached commit_count in share
  Fix so that ndb_util_thread uses milliseconds "sleeps"
  Changed so that ndb_commit_count uses the commit_count from share if available
sql/ha_ndbcluster.h:
  Add commit_count_lock to NBD_SHARE, use for detecting simultaneous attempts to update commit_count
  Add list of tables changed by transaction to Thd_ndb
  Change check_ndb_connection to take thd as argument, use current_thd as default
  Added m_rows_changed variable to keep track of if this handler has modified any records within the transaction
sql/set_var.cc:
  Change format of code
  Sort sys__ variables in aplha order