" Torque will be proportional to the phase angle between the stator field and the rotor position "
If the underlying technology is a 3 phase brushless motor of some form, you'll find that torque can also be varied by changing the amount of current flowing, as well as adjusting the relative angles between the rotor's position and the angle of the field the current in the stator is making.
You've described this thing as a servo, which implies it comes integrated with the motor, feedback sensor (encoder of some sort) and control logic all together. In that case its own datasheet should tell you everything about the commands to run it, how much torque it can supply... If the controller it comes with is reasonably capable I'd be very surprised if it couldn't run right down to a very low speed indeed, like something so slow you start to really see cogging-like jumps as it turns. The exception would be if there is reliance on sensorles measurement (measuring induced voltages in the undriven coil of the motor so as to work out which location it is rushing past as a consequence of the current in the other two), but that would be a very strange choice for a servo, the whole idea of a servo afterall is being able to command it to go to particular positions.
I've you've just bought the motor and feedback sensor, and it doesn't come with integrated control logic, you'd either look for an appropriate controller device, or design your own. I've been designing, for a much smaller 3 phase BLDC style motor, recently such a controller board and the overall principle for operation in a servo-like mode is quite simple.
Each phase of the motor is given a PWM signal between 0% and 100% (0 to 256 in my case), you use a microcontroller running logic of the form:
Apwm = beta*1.14*(sin((float)angle*PI/180.0)*127.5+25.5*sin((float)3*angle*PI/180.0))+127.5;
Bpwm = beta*1.14*(sin((float)(angle+120)*PI/180.0)*127.5+25.5*sin((float)3*(angle+120)*PI/180.0))+127.5;
Cpwm = beta*1.14*(sin((float)(angle+240)*PI/180.0)*127.5+25.5*sin((float)3*(angle+240)*PI/180.0))+127.5;
The added extra term (+25.5...)gives you SVM (space vector modulation) which lets you get the absolute maximum from a given supply voltage, but this can be neglected and just the first term used if you never need to put power in to the motor near to the maximum your driver can supply.
You vary "beta" to change the (approximate, beta isn't linear with it, but it is monotonic, you can calibrate for it in a finished system, or use current measurement of total current drawn for further feedback) level of torque. Beta is always in the 0 to 1 range. If beta is particularly low then even the phase getting the maximum current at a given time gets only a very quick burst of power to it (during which that leg of the motor is connected to the supply voltage, and the current flowing rises with a slope proportional to supply-voltage/motor_inductance), then the half-brdge which powers that phase conducts between the motor's leg and ground instead and the current recirculates, buck-converter principles.
Then the angle (here in degrees, 0 to 360, and it is an electrical angle, not shaft angle, so you have multiple electric rotations per shaft angle deending on your motor's pole pair count) you set as you move, you control the speed by having your software increment of decrement this angle as fast as you want.
With that working you can then use feedback to know where the rotor is, vary that beta value, if you need to, to reduce the current when not much is needed for the torque demanded by a load (when the rotor angle is lagging the stator field by a "fair bit" (amount to be determined by testing and encoder precision) less than 90 degrees) and increase it whenever torque demand rises (rotor angle lags stator by a "fair bit" more than 90 elec degrees), and even compensate for skipped "steps" if they occur.
This principle can work at any speed you like, and any level of torque, up to the maximum torque the motor can give at the highest level of current it is rated not to overheat at*. Yours sounds like a big industrial 3 phase motor? So you'd need much fancier half-bridge circuits than I used, but the "beta" and "angle" PWM method would still work.
*ofcourse, using a high power motor won't help your "we want a high powered motor at slow speed rather than a low powered one with a gearbox" situation if this maximum current, independent of maximum power, can't give you enough shaft torque. A motor at maximum power is going to be both drawing the maximum current, but also using power to overcome back-emf voltage caused by running at high speed, at lower speeds you cannot draw maximum power without drawing more than maximum current, which via I^2 R heating would damage the motor, given "enough" (seconds, minutes, hours... depends on the motor, how hot the surrounding environment is, how much you over-current by...) time even though the motor was well below maximum power. If the higher power motor's maximum current rating, multiplied by its torque per current value (for cheap motors that is 8.3/kV with the kV rating in rpm/V and the torque-per-amp in Nm/A ), won't give you enough torque, you'll need a gearbox nonetheless.