First, the file: which frame should I be looking at? What should radius be doing on that frame? I think I know the root of your problems anyways though.
TL;DR: Euler angle axes don’t work the way you think they do. They’re not independent of each other. There’s not really any such thing as affecting an XYZ Euler angle X axis without also affecting what you perceive as the Z axis. Use other constraints (damped track and locked track) to control the orientation of bones instead.
In this particular case, it’s exactly what I was saying about not using limit rotation and copy rotation.
Limit rotation acts on XYZ Eulers. It decomposes the transformation of the bone, which is a 4x4 matrix, into a XYZ Euler rotation, clamps the explicit numbers in that rotation, and then recomposes a new 4x4 matrix out of its new triplet. It’s a bunch of math. Not necessarily simple, and it doesn’t match your intuition, but it’s just math, and you can trace it out if you want.
The problem is that the way that Eulers work, your Z rotation is not independent of your X/Y rotation. Changing your X/Y rotation changes what you’re thinking of as your Z rotation. Euler rotations are ordered rotations, and the axes aren’t actually independent of each other. This also screws up your copy rotation constraint, which is copying only the Y axis rotation of its target, but the Y axis rotation of its target depends on the X axis rotation of the target (because it is an XYZ Euler: X, then Y, then Z.)
Typically, when I want to do something like this, I’m going to do it with damped track and locked track constraints. Think about the orientation of your bone as 2 things, rather than 3. First, the point that the tail of the bone should be pointing at. (Damped track Y that point.) Second, the point that its X (or Z) axis should be pointing at. (Locked track, lock Y track X, that point.) The third axis is fully defined from those two things.
In your structure, if you just damped track some point with Radius, it won’t really do anything, because it’s parented to the control-- it’s already inheriting the full rotation of Forearm, so it’s already pointing where you want. You may want to parent it, instead, to Humerus (or whatever.) If you don’t want a bone to copy all rotation, don’t try to use a limit rotation-- instead, parent it to something else so that it never gets that rotation in the first place.
In many situations, only one of these constraints is necessary to get the behavior you want. A damped track Y is the shortest path rotation to the direction to which you want the Y axis to point.
Copy rotation has some options to control “order”. It’s usually possible to get what you want with that, but it requires really understanding how Euler angles work-- how they’re not actually 3 independent axes. Limit rotation does not have any “order” field, however. It always limits bones as if they were XYZ Eulers.
Often, anatomical terms are not the best for thinking about these problems. When anatomy texts are talking about “pronation”, they’re talking about pronation of the hand. But the hand is dependent on the radius. We can’t talk about the radius in terms of the hand because of that dependency. There’s a lot of translation we have to do with anatomical terms. (There are other times that anatomical terms are very useful though, like for naming features.)
Note that this whole thing doesn’t make a lot of sense treating Forearm as the ulna, since ulna doesn’t inherit the full range of transforms we can give the forearm, and radius instead inherits some of them. Instead, we should have deforming bones ulna and radius, and controlling bone forearm which deforms nothing, with both ulna and radius dependent on the control.