An UPDATE_ROW binlog event contains 2 versions of the updated
row: a Before Image which has the row before the update,
and an After Image of the row after the update. The Before
and After Image do not always include all the columns in the
row. The binlog format decides which columns will be included.

There is a requirement in ndbcluster to reduce the size of the
binlog event while retaining the UPDATE_ROW format of
BeforeImage/AfterImage. This can be done by writing only
primary key columns from the Before Image and only modified
columns from the After Image. However, the existing binlog
formats do not support different sets of columns in the Before
and After Images.

The --binlog-format=MINIMAL option is not usable for ndbcluster
because enabling the binlog format code breaks the
--ndb-log-updated-only option in ndb replication.

This patch implements 2 new ndb binlog formats for UPDATE_ROW
where the Before Image and After Image are minimised to remove
duplicate information.

UPDATED_ONLY_USE_UPDATE_MINIMAL: log update as UPDATE_ROW, log
only primary key columns in Before Image and only updated columns
in After Image

FULL_USE_UPDATE_MINIMAL: log update as UPDATE_ROW, log only
primary key columns in Before Image and all non primary key
columns in After Image

This means that each column is contained only once in each event,
bringing efficiency close to that achieved by the variants logging
with WRITE_ROW events.

The UPDATED_ONLY_USE_UPDATE_MINIMAL and FULL_USE_UPDATE_MINIMAL
formats can be used with the conflict resolution algorithms
NDB$EPOCH(), NDB$EPOCH2() and NDB$EPOCH_TRANS(), because these
do not require the 'Before' values of non-primary-key-columns.
Conflict resolution algorithms like NDB$MAX() and NDB$OLD()
require the old values of non-PK columns, and will not work
correctly with these new variants.