Yes, it is a sub-set of your "smooth" category. I proposed it as being useful due to the fact that it can improve smoothing results (as opposed to improving smoothing performance, which is what FLAT and SMOOTH_ATOMIC do). In my example, without semi-flat (or something similar), the normals at the top of the rect3 would be pulled half-way up to meet the normals on the cylinder. This will cause that bottom face to disappear visually into the cyllinder, when in fact that face is flat in the actual part.
I'm not sure that adding a fourth category to your proposal is the best way to handle this case, but it seemed like it would be straightforward to do so.
I'm not sure that adding a fourth category to your proposal is the best way to handle this case, but it seemed like it would be straightforward to do so.