There are two synchronization cases to consider:

1. A fresh sync where you need to copy the entire table
2. An update, where you just want the rows that have changed or been deleted

Ideally the same code would work for both cases - a fresh sync just starts at the beginning.

I want to take advantage of the index on the updated_ms column - and use it for efficient keyset pagination, to handle arbitrarily large tables in limited memory.

Will eventually want this to work over a network eg via a JSON API provided by Datasette.