Redundant use of recordset / allow streaming data iterators

[brendanheywood]

The database source uses a recordset but then defeats the purpose by grinding through it and copying it all into an in memory array:

https://github.com/catalyst/moodle-tool_etl/blob/master/classes/source/source_database.php#L59-L66

But ultimately we need to be able to handle massive data so we need to change the API to allow extract() to return an iterator, and then each source could return a db recordset, or it could open a csv and stream through it line by line etc