That isn't the basic problem, so much as the failure to note that the two-p ole all-pass is always going to be an attenuator.
Rich S. has dug the general solution out of the Analog Dialogue library, wh ich shows that there is a range of component values which give a range of a ttenuations, which is probably what you would like to have published, if yo u had come across it (which I certainly hadn't).