A rational model of the Dunning–Kruger effect supports insensitivity to evidence in low performers