MYSQL SERVER RESTART

When there is a MySQL Server shutdown or a cluster failure during a
schema distribution, where the coordinator is waiting for the
participants, the schema distribution is aborted halfway. The
ndb_schema_result rows, so far received by the coordinator, cannot be
cleaned up at the end of a cluster failure, as the cluster is already
down. And, during a server shutdown, even if the rows are cleaned up
before the binlog thread exits, there is always a chance that a slow
participant could reply after the coordinator has shutdown. So such rows
related to the schema operation, which has now been aborted, end up left
behind. There is a chance that these rows might have a conflict with a
future reply from a participant if a DDL with the same schema op id
originates from a client with the same node id.

The solution is to clear all such rows on ndb_schema_result during the
ndbcluster binlog setup. Since the binlog thread will be holding the GSL
during the setup, it is assured that there are no DDL distributions in
progress and any rows remaining in ndb_schema_result table are already
obsolete.

Changes:
- Added a method delete_all_rows() to the Ndb_util_table class which is
  now called during the binlog setup to clear all the rows in
  ndb_schema_result table.
- Added a new function ndb_table_scan_and_delete_rows() that scans the
  rows of a given table and, deletes them. Rows can also be optionally
  filtered out by defining a NdbScanFilter.
- Updated Ndb_schema_event_handler's remove_schema_result_rows() method
  to use the new function mentioned above.
- Added a new method get_column_num() to the Ndb_util_class which
  returns the column number of a given column within the Ndb table.
- Updated the retry functions in ndb_retry.h to take in a constant
  reference to the std::function.

Change-Id: Ia96365fd60a84f6257ad0acbcef60d0e168e1995