Took the step 0 for this project. I needed basic proof that I can pull this off.  Basically a minimal implementation that can read a KiCAD board, process it and create something that tells me this will work. I am developing this first and foremost to work for me. Here is AtiVEGA with the results of today's work:

This is a screenshot of FreeCAD. Essentially overlays two models - a STEP file exported from KiCAD and a CSG file exported from OpenSCAD.  The CSG file represents the 3D printed parts. that will serve as the holders for the 3D printed components. These are shown in gray with red lines. As you can see, they sit damn well ! How well ? Here is a top view of the left hand top corner of the PCB. If you zoom in, you will see a tiny sliver of green between the 3D printed part model and the through hole components. That gap is 0.1 mm and the thickness of the shell itself is 1.2 mm.
Let's zoom in a bit more
Sliced it - layer height 0.35mm to make the Z lines visible !
My hands are itching to print, but my mind tells me to wait. To check the tolerance and fit, I'll have to print anyways. I use cut headers a lot of times, as the commonly available headers are 2x40. That will foil my tolerance tests. But in general this looks neat. This approach will work. I am convinced.

How does this work ? In short, my script iterates over all the TH components. Loads their 3D models. And transforms them to their final position on the board (notionally at least!).  The simplest method to hold the TH components in place is with a simple shell, as shown above.  But how do we compute the shell ?

A simple method (KISS for now) projects the 3D model of the component (e.g. 2x20 berg header) onto 2D space - say the top layer. Just drop the Z components, that will do. As a result, you get a whole lot of (x,y)s - essentially points that represent the 3D model.  Now, you take a convex hull.  This is guaranteed to encompass all points.  It also has a nice property that it wont "hug" the model in its nooks and crannies - e.g. the gaps between the 2.54 mm spaced plastic housing. The below picture in wireframe should make this clear.

Will this approach work for every component ? Definitely not. But does it have to - again not, at least initially. I'll probably include simple overrides at the component and footprint level so that the generic nature of the tool is preserved. That way, anyone using it can even make use of most of the generic functionality - and include their own overrides flexibly.

Now that step 0 is done - here are my next steps:

1. close the holes with caps so that the components can slide in and sit (now, won't that be satisfying!). This is simple, and I can easily accommodate components of various heights.
2. add a frame on which the PCB will sit. Need to dig into more "pcbnew" for this, and/or lookup stencilframer.
3. merge the caps into the frame. A simple slab matching a bit more than the PCB shape will do, but it will be fun to come up with ways to save plastic in the 3D print ! Premature optimization is the root of all evil, says Knuth.  But yet, me starts the think!