Migration SQL written for PostgreSQL, H2 and SQLite, rewritten for MySQL.

The scaffolder writes one migration for every database (BOU-569). MySQL has
no UUID or zoned TIMESTAMP type, takes no IF NOT EXISTS on an index, and
ignores a REFERENCES written on a column. Only that is rewritten, so SQL
written for MySQL passes through as it is.

Pure functions for SQL query building and transformation.

All functions in this namespace are pure - they take data and return data
without side effects. No I/O, logging, or state mutation.

Pure logic for database seeding.

A seed file is EDN, in either of two shapes.

A map of table -> rows, for the simple case:

  {:tasks [{:title "Try the admin UI" :done false}
           {:title "Read AGENTS.md"   :done true}]}

Or a vector of [table rows] pairs, which is ordered:

  [[:users [{:email "admin@example.com"}]]
   [:tasks [{:title "Owned by that user" :user-id 1}]]]

Insert order matters as soon as one table references another, and EDN maps
only preserve their written order up to 8 entries — a 9th turns the literal
into a PersistentHashMap and the order becomes hash order. Rather than let a
seed file quietly start inserting children before parents once it grows, a
map larger than that is rejected with a pointer to the vector form.

Table and column names are written in kebab-case, like the rest of the
codebase; the conversion to snake_case happens here, at the point where the
data becomes a persistence concern.

Everything in this namespace is pure. Validation returns typed error values
rather than throwing — the shell decides how to present them.