That's because in the current version, embarkability is only checked before scanning the region at all. If you visit a new tile on the world map, the game automatically places the embark rectangle in a position that is embarkable by the player IF such a position exists. I use this fact as a quick way to determine whether it makes sense to do the scan at all. You can observe this on mountains and oceans, where the script passes almost 4 times as fast over tiles as it does in the plains. However, the script does NOT yet check if the found features can be fitted in an embark rectangle that is actually embarkable.
Yesterday, I have coded the beginning of the second-level scan, which will include stone layers. It interactively forms an embark rectangle according to the settings in the header of the script (max_area and max_sidelength) and positions it over all found features. And in this routine, I am doing the REAL embarkability test. It already works and it's quite funny to look at. There's also a lot of optimization potential there, I kinda took the quickest-to-code route, because I wanted to proof the concept.
I wasn't able to tidy everything up, though, so this is why you don't get the download yet. Maybe I have a chance to visit my apartment tomorrow noon, or perhaps I can use my father's computer again.