My guess would be that they changed the way that keyboard events are processed, so that killing keydown
also automatically kills keypress.  It would be noted in a changelog or the "What changed" document if so.
An experiment that you could do would be to alter key-kill-mode so that it doesn't kill the keydown event,
and only kills the keyup event.  This would undermine its purpose, but it would at least tell you whether
this theory is correct.