Expand description
The fse migrate flow: parse the tables folder, diff against the
snapshot, write a plain sqlx migration, then apply everything pending.
The snapshot is updated at generation time: it records the schema the
generated migrations produce, while the database’s _sqlx_migrations
table tracks what has been applied. So an aborted apply, or the
edit-the-TODO-then-rerun flow, never generates the same migration twice —
rerunning just applies what is pending.