Handle primary keys that aren't auto-populated

We currently presume that PKs can be ignored and will be auto-populated by the DB. Although this often works, there are plenty of cases (see, for example, the OMOP domain.domain_id column, which is a VARCHAR) where we need to specify one.

We may need to inspect each PK column to work out whether it will be auto populated and, if not, choose a *unique* random value for it.

Start by tracing how primary-key columns are currently handled during synthetic data generation. Use the OMOP domain.domain_id VARCHAR example to examine the distinction between auto-populated and explicitly required keys. Done means non-auto-populated primary keys receive unique values while keys populated by the database continue to work.

Written by the indexing model from the issue text.