As was run on production during migration.

The migration looped over all rows (node, muser) from versions,
and update muser in each one.

However, the tuples (node, muser) are not unique in the table,
and more important, the muser values are much fewer than the nodes
(since there are by definition much more users than files).

Instead, we now loop over the distinct muser values,
and issue an update statement for each one (which updates much more
than one row, one for every version this muser has produced)