2HandedAfforder: Learning Precise Actionable Bimanual Affordances from Human Videos