Can this be done, probing to DXF/STL?

I think that the implementation of this would depend on the type of probe being used. Strain gauge probes can give extra information about the direction of touch, and I think this is what the part-tracing CMMs use, rather than a simple touch-probe.

Part tracing with a touch-probe has to guess what angle the probe is touching the part at.
I imagine that you would simply assume that the current point is an extension of the line through the last two points during the scanning process, (to calculate the next probe-start position) then calculate a better guess a the real tangent once all the points are gathered.

Psudo-code:

Pick a path direction, let's assume anticlockwise.
Probe from start point in the direction suggested by the operator.
return to start point
[1]probe at high speed in the direction of the last segment anticipating a no-touch.
If that move ends with a touch, back off along the previous path, turn the probing direction 10(?) degrees clockwise, try again.
if that move ends in no-touch, probe towards a line 10 degrees anticlockwise of the last segment
if that hits, retract a little and loop back to [1]
if that does not hit, retract to a point on the last segment, rotate the path 10 degrees, repeat...

You need to consider the situation where the probe might get stuck in a probe-sized hole. So I think that every move needs to be a probe move.

G-code can log data to file, so the whole thing can be a standalone G-code program.
linuxcnc.org/docs/2.7/html/gcode/overview.html#_probe_logging

Probe-logging (15) will probably put excess data in the file, so normal logging (16) of only the moves that expect to find a part would be better.