the second time the slot is synced, there is no prediction, so the slot update isn't associated with a request anymore. This causes subsequent requests in the same packet to fail, since the dependency request ID isn't associated with the slot anymore.
This change fixes the problem by only allowing the backing inventory to trigger a call to DelegateInventory->on*Change(). While we could have removed and re-added the listener instead, this way is safer since it doesn't assume that the backing inventory won't modify the given item in setItem().
closes#5692
