The Many Faces of Optimal Weak-to-Strong Learning