Voltage of the output stage (motor windings) may be sensed to provide so called "sensorless" rotor angle estimation. If this scheme is not in use, then voltage measurement isn't needed either. A servo motor, i.e., one with positional feedback, obviously won't be using rotor angle estimation, so doesn't need phase voltage measurement.
Voltage of the input DC bus is important to measure, for cutting regeneration in case the energy can't go anywhere upstream and DC link bus voltage is rising towards dangerous levels. Other reasons might be if some control loop benefits from feedforward from the input voltage.
Doctorandus_P is othewise correct but the name for sensorless commutation is not "FOC", FOC is something completely different. Rotor angle (estimated or directly measured) is an input parameter for FOC, though.
The key of understanding of FOC is not that it provides optimal 90 degree phase shift. This can be provided in simple, totally trivial ways. The reason to use FOC is that it makes the PI control loops (input: current measurement; output: PWM duty cycle), "rotate" in sync with the rotor, making the delay from the PI loop insignificant. Without FOC, you would have three simple PI loops to control phase currents, but because the setpoints would change sinusoidally and rapidly, there would be lag in actual current. All FOC means, is we use a single PI loop and rotate the inputs (phase currents) in sync with the rotor angle, then rotate the outputs (PWM setpoints) back. Then, the current setpoint is a straight line, only changed when you want to change torque generated. The sinusoidal output currents are simply generated by the sin() and cos() functions when calculating standard 2D vector rotation in the FOC implementation.
At least for me, trying to understand what FOC is was originally a bit hard because no one explained the real function / reason properly, it was either advertisements by people who have no idea what they are talking about, or obfuscated mathematical explanation about the implementation of the formulae, which I also wasn't interested about before I know why. Besides, in reality, the math is very simple, such as standard 2D vector rotation. In the context of FOC, this standard formula, familiar from computer graphics, has been just given a separate name for obfuscation.