Let's try out this pattern: feature branches should collect new SQL
statements in a branch.sql file, and if this exists we'll apply it
during makedb.sh. That should avoid some conflicts when merging from
master out to feature branches, as well as making it clearer that
branch.sql is a work in progress.