Sokol, the orientation of his hall sensor is different than the sensor in your illustration. For his configuration, he is correct to locate the sensor directly on the axis of rotation.

Volans, if you're already getting enough output change to swing through 10 bits of precision, you should be good to go. The extra 2 bits sounds good on paper, but doesn't really get you much benefit. As it is, 10 bits usable resolution gives you ~50 steps for every inch of stick movement. My hand moves more than that due to my pulse alone.

If you don't like the larger movements introduced by a long stick, you could simply calibrate using less movement. You've got the resolution available, and would probably not notice any decreased performance even if you reduced your throw to +/- 6° (still ~8 bits usable resolution, typical of most sticks on the market).