I still think the issue is related to your having used open loop Sensorless Vector Control, and the drive not being able to ensure there was enough shaft torque. So the drive was putting out torque, just not enough to keep the load from falling. But that only became evident when that rapid direction change took place. By adding the encoder feedback, the drive now knows what's happening in the motor.
One issue I see happen to people is that they witness an SVC drive working on a hoist, and assume they all work on all hoists, which is the dangerous assumption. They work when there is a mechanical issue that will not allow the load to drop anyway, such as a right angle worm gear driven hoist, or a conical motor like on a Demag hoist that will not let go of the rotor until AFTER the motor has already developed sufficient torque to hold the load.