The Arkhangel’skiĭ–Tall problem: a consistent counterexample