I think one could read the directive to mean "tie [all ground symbols in this figure] to Vref/2"; which reduces the drop along the R2R chain by half, which is still something, and, I'd have to think about it but I think that would just give 0...Vref/2 range, which is indeed positive, if half the range advertised otherwise.
Not that I would expect this reading to be the obvious interpretation, and some clarification would still be deserved.
Else, if changing just the one... that still works, doesn't it? The resistor divider doesn't care at all where you connect it, as long as it's to a voltage source. And then it would get either Vref/2 (switch up) or GND (switch down), both fixed voltages. Charge injection would stink, as the switches are going between Vref/2 and GND, and probably error would increase (due to analog switch Rds(on) varying with Vcm). Which is easily fixed by moving the "dump" rail to Vref/2.
And for sure, if the divider were moved to Vref/2 (top, "dump") and GND (bottom), it would be strictly below V(+in) = Vref/2, so the output would be strictly above, i.e. Vref/2...Vref (or some ratio thereof). Which is also positive, but not zero-inclusive, which kind of sucks.
Tim