Your equation for current limit is incorrect. According the Maxim datasheet, the expression in the brackets is (R4+R3), not (R2+R3). That may or may not have been confusing you, as R2 is closely associated with the PFET.
With the correct expression, we can see that the circuit is servoing the drop across Rsense to be proportional (equal to 0.4x) to a reference voltage on R3, the Vset voltage.
The reason that this is an 'inrush current limiter' rather than a 'current source', is because the reference voltage and therefore the output current depends on VBAT, which in most applications can be expected to vary, perhaps over a 2:1 range, not accurate enough for most people's idea of a 'current source', but good enough to limit current inrush.
You can implement a current source in closed loop mode, as there is an ADC in there (there's an ADC in there!!!) to measure what you have, and adjust it.
How accurate do you need your current source? The voltage on the SET pin needs to be known (either by measurement, or by simply setting it) to better than this accuracy, to allow for errors in Rsense, the 2.5x gain factor, amplifier offsets to have their chew at the accuracy budget. I haven't read the data sheet in enough detail yet to see what sort of accuracy could be expected even if vSET was known exactly. Maybe they add up to a large figure, perhaps that's why it's an inrush limiter rather than a current source? I'm a little disturbed that when reading the SET voltage, they quote the typical error as 0.2% and the maximum error as 5% !!!