Have you tried to georeference the images and then process to orthomosaic? That could be tedious but would add coordinate information.
Alternatively, you could take ground control at obvious points of strong geometry that still exist in the modern day (manhole cover, road intersections, etc). The images could still be combined using structure from motion, and referenced with the GCPs.
I am not a drone2map user - still using Metashape and OpenDroneMap. Curious if you are able to find a solution to this one.