fix: add startup workaround for trainwreck query bug

not at all clear why, but c621e21cd4b4a2a408d8a32ace0b34afe6fdf526 has
introduced some strange bug where the "first" trainwreck query will
raise an error and yet subsequent queries work fine.

so now in the startup sequence we simply force a query, which for
unknown reasons does *not* raise any error, and will somehow magically
prevent the error when a "normal" trainwreck query happens later.

AFAIK this is a "benign" operation and causes no real side effects
other than preventing the strange error. fingers crossed, this is
fine and will not cause further problems...??!

fix: refactor code so most things work without sqlalchemy

also refactor tests to help ensure that remains true