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: change import for wuttjamaican base model

must avoid importing the actual model subpkg from wuttjamaican because
it can interfere with sqlalchemy-continuum versioning

fix: move "record changes" global hook to startup()

this silly thing shouldn't even exist but can't remove until i'm
certain it's not in use in the wild.

but at least, it shouldn't be triggered when a config object is
instantiated, it needs to be deferred a little. should not affect app
runtime per se but more helpful for tests.

fix: cleanup old code for "record changes" session feature

had some tests failing, and one thing led to another.. apparenty we
still had support for some pretty old sqlalchemy so no need to keep
that around.

fix: hopefully fix startup continuum bug per 'active_history' models

only have one model in the wild (SpecialOrder) which uses
active_history and it probably shouldn't really.. it was written a
long time ago and would need some refactoring.

but for now we just need things to work, and this seems to do it
although i have no idea why