This question makes me go back to my college years when a teacher was trying to "explain" operational amplifiers to me through definitions. When she told me "the input offset voltage (Vios) is the voltage that makes the output voltage be zero", I could not imagine it...
So, the output voltage Vos of the single op-amp is not caused by some input voltage but rather by something non-electric - the difference between the parameters (asymmetry) of the two parts of the differential circuit... and we cannot measure it directly. We can measure this imperfection indirectly by neutralizing its impact to the output voltage through an equivalent input voltage Vios.
For this purpose, we ground the circuit inputs, add the compensating voltage in series to the differential input voltage, and vary it until zero the output voltage. We can connect the compensating voltage source in series either to the inverting or non-inverting input.
In the discussed inverting configuration, we do the opposite trick - we add voltage equal to Vios (according to the op-amp specifications) in series to the differential input voltage with the purpose to emulate Vios. It acts as the only input voltage that is compensated by the voltage at the output of the negative feedback network (the voltage divider R1-Rf). We discern the well-known non-inverting configuration (a little odd if the floating compensating source is connected to the inverting input). Hence the expression for the resulting V0 = (1+ Rf/R1) * Vos.
Indeed, it is quite strange to consider an inverting amplifier as non-inverting... and maybe this is the reason for the author's mistake above.
Again, we can connect the compensating voltage source in series either to the inverting or non-inverting input... but the latter is more suitable since the source will be grounded.
The trick to connect a floating input voltage source in series to the op-amp differential input is old and it is shown in the famous Sheingold's paper (Fig. 4 on Page 7).