Hi Ben,
I took so far the whole morning to explore if I’m not too pessimistic with my opinion that I gained over the years about it. Here is my conclusion:
Your questions would lead to think about a proper Pole set up and limit the Joints>Kinetics to get a mechanical rig. I have not seen a set up with so many axis freedom/constrain and a single IK, with no extra rigging, that worked.
https://help.maxon.net/us/#TCAIK-ID_TAGPROPERTIES
I can see that this comes close to explain the underlying problem
https://www.cineversity.com/vidplaytut/creating_a_no-flip_knee_rig
Brainstorming: If the core knee could be defined, the positions of that joint set up could be used. All parts need then to get their orientation from Target setups. Again, I feel even suggesting it uncomfortable. If I interpret your set up wish correctly, I count seven cases where the joints need freedom, which needs to be solved, besides eleven that need to be constrained to zero movements. The numbers might vary by a few, dependent on the target.
As a side note, the Goal might not work alone, more input might be needed
I’m currently not convinced that the time spent to build a rig with all needed parts around it, will give you in all cases a stable result, as well as, that the invested time is not larger than to animated it manually.
However, it is undoubtedly a great exploration to see where the limits are, and perhaps you develop something useful anyway. Enjoy the time developing it.
This would be the maximal control that a single Goal can direct. Anything else needs more input when the lower or/and upper arm follows (or not). That differentiation will make the set up more complex and closer to the manual set up.
https://www.amazon.com/clouddrive/share/0NXPnxjTIzJA6SiJdkbKMpvSyLUcVEVvlgRbtKNDYo8
My best wishes.