The sensor datasheet tells us each voltage divider has a resistance of 438 ohms (nominally). For a sensible power dissipation in each divider, you're going to want to keep the excitation voltage low. At 5v, each divider is going to consume around 11 mA and dissipate about 60 mW. If you supply the dividers with the same voltage you use as the ADC reference, then the entire system becomes ratiometric, so you don't need to worry about the stability / variance of your voltage reference.
In their suggested schematic, the 10k resistor and 0.22uf capacitor provide a low pass filter, in this case, with a 3bd point at 70Hz, which would suggest you need to sample each channel at around 150 Hz to avoid aliasing.
The higher the resolution of your adc, the higher the resolution of your system (for any given level of noise). Something like an arduino with a 10b adc, has 1024 steps, so each of those voltages can be decimated into 1024 descrete steps, so:
1) for the position sensors, that have a FSD over 180 degrees, that a theoretical max resolution of 0.18 degrees
2) for the torque sensors, that have a 10 degree range of 80% of FSD, that's a theoretical max resolution of 0.122 degrees