I'drather use an AC optocouple, H11AA1 is the one I've used to sense voltage on a load, rather than current on a line. Then you just add 2 diodes in parallel each direction (one wouldn't cut the 1.5V forward voltage on the coupler).
Same circuit would work using a conventional optocoupler, 2 diodes in parallel, only one in the oposite direction (if the offset isn't a problem) The only problem is the smoothing cap, which I think for any case should be put in the other side, Time constant long enough, 3 cycles? 50Hz for the conventional optocoupler, 100Hz for the AC optocoupler. You can get away with a much smaller cap. I used 47µF IIRC.
For higher currents a shunt could be used, we are stuck with 2V now so 20A would make 40W anyway. The shunt needs to handle all that power, and the sensor will end ignoring lower currents which would be detected if just using beefier diodes.
In any case, how much does the components on this costs, and how much does a proper hall current sensor costs? The drop/power is not even considered and you can measure quite well how much it is. The most problematic part I guess is to handle the analog output but in the worst case a comparator does the trick, with easily adjustable sensibility.
JS