Reminders to myself or whoever else ever looks in here...

SQLite needs some heavy refactoring, the subroutines are becoming too complex to understand easily.
MySQL needs implicit FK support, I think.
the base/use injection stuff needs error checking/reporting, and some testing

Consider:
   If local column is UNIQUE or PK, use has_one() for relation?
   Re-scan relations/tables after initial relation setup to find ->many_to_many() relations to be set up?
   Check NULLability of columns involved in the relationship, which might suggest a more optimal non-default -join-type?

...