User story
As a developer I would like to make the following improvements to the 'filtered positive rule change migration'.
- Accept start/end time inputs - so that it can be run in batches and more easily re-started if it goes wrong
- Remove the DART pending plates logic in favour of just not cherrypicked samples - because currently, it won't update samples that are not in DART at all, and we might want to go back and pick old samples.
- Use the more performant MLWH update method - because it's probably needed to make it fast enough.
Who are the primary contacts for this story
Katy, Philip
Acceptance criteria
See user story