@Bug.Robotics , I worked with Karina on this issue yesterday, and I understand better now what is going on, and I think I can explain it. I still agree that the result is undesirable, but thought you'd appreciate a detailed explanation.
Projecting a body into a sketch is something we call "silhouette projection" internally. The term is pretty self-explanatory - we want to take the apparent edge of the model as seen from the direction of the sketch normal. In the case of this model, the side faces of the model are all parallel to the sketch normal. When possible, Fusion looks for edges that it can project to represent the silhouette (there are some cases, like a sphere, where there are no edges, so we have to generate the curve differently, but if there are edges, these are used first). So, there are two choices for each edge when projecting into the sketch. Let's look at a couple of these edges. The first one looks like this:

I've selected the projected curve and the two edges which are equal candidates for the representative silhouette projection. Both of those are circular edges, so the result is a circle in the sketch.
The second set, though, is where the issue occurs:

either of those edges would work, and are equivalent as far as silhouette representatives. But, both of these are actually spline edges in the model, as you noted, because of the chamfer. So, the result is also a spline (even though, technically it could be a circle, but that is not how the modeler works)
The third one is similar as well:

both edges are equivalent projections, and both are spline edges because of the chamfer.
The interesting case is when we only chamfer one of the edges:

again, there are two candidate edge sets here (one single edge, the other a chain of two) that could be used. But, in this case, Fusion picks the front one, which does result in a circle.
The last thing is another type of workaround you can use: If you put the sketch before the chamfer, then this is much cleaner:

because you can then project the silhouette without the geometric complexity that the chamfer adds.
Jeff Strater
Engineering Director