I think it's sort of a trick question. They can have ACW on the voice segments within the callback, but the callback itself doesn't have ACW because you can spend any portion of the callback sitting idle already.
A callback is a media type, it can be due to an initial voice call but it doesn't have to be. an outbound campaign uses callback as it's media type and has no initial voice segment and other scenarios can utilize callback without their being an initial voice leg. the nomenclature is a bit misleading.
Wrap up isn't mandatory for voice calls either. Even in your scenario a queue with it optional could have zero wrap up.
Because call back and voice are distinct media types they have distinct segments and metrics and just because the voice segments are related to a callback their metrics and segments remain bound a subset of themselves, not their callback cousin.
If you in terms of the conversation tree;
Conversation
Participants
Sessions
Segment
Metrics
Even on a callback that is directly due to a voice, you'd have three distinct sessions there, each with their own segments and metrics. one voice, one callback, and then another voice that's is temporally overlapped by the callback but remains entirely self contained within the tree as a sibling not a child. The wrap ups would both be segments in the two voice sessions. The Callback session will have zero.