DRV8323 is very nice indeed! I found that the current sense for torque control works pretty well on S3. All my analog channels are on ADC1 which helps, but I do have to call an analogRead() once at the start on those channels for it to not crash while initializing FOC. Once I do this, it works well, though.
I did a lot of testing with an industrial geared 1pp BLDC with halls / encoder and some open loop testing on some smaller outrunner motors. The only issue I can see for much smaller motors is the current sensing resolution at low currents. The shunt amplifier gain can be doubled by a solderjumper in that case, which should help a lot.