Revisiting the Taylor-Culick approximation. II. Retraction of a viscous sheet