The job of R2 and R3 is to create a voltage that is roughly half the supply. Lower values give you lower impedance of the output voltage, but also increase current. In this case, the voltage only needs to drive a opamp input, which is high impedance. The 50 kΩ output of the divider is plenty low enough, and 45 µA of current was deemed acceptable.
C1 reduces noise on this voltage reference. The noise can come from the power supply directly, and can be picked up from other signals by stray capacitive coupling. With 100 nF and the 50 kΩ impedance, the low pass filter rolloff frequency is 32 Hz.
The choice of 100 nF was probably a knee-jerk reaction. That seems to be a common value to grab as a substitute for actually thinking about the tradeoffs. I would probably have used 1 µF. The downside of too much capacitance is that the circuit takes too long to settle after initial turn-on. With 3.2 Hz low pass filter, the settling should still be fast enough on a human scale. If this circuit is switched by machine, which then wants it to be ready after a short time, then 100 nF could be better. That doesn't seem to be what is going on there, though.