Hi Marko,
OK, it is not dead-end for everyone, because the highway=pedestrian area (plaza) 35826685 is connected to the driveway. We may route pedestrians on the edges of the highway=pedestrian area. We might route bicycles too (AFAIU, it is allowed in Finland with maxspeed=20, but not allowed in Germany).
It is the same thing if you had a oneway highway=service that ends in a path or steps. It is not dead-end for pedestrians, but would be a dead-end for cars.
I wrote some months ago about a possible project to highlight dead-ends on the map, but nobody replied. I think it should be a set of 'dead-end layers', one for each mode of routing. Then you could easily survey all those nearby suspectable cases when going for a walk or a bicycle trip.
Pedestrians can use steps, cannot legally use highway=cycleway. For bicycles it is vice versa. Both can use paths and most ways except motorways. Cars can use all ways except those meant for bicycles and pedestrians. There are really multiple routing graphs to consider if you want to check for dead-ends properly.
I agree that it would be better to check each road network separately, but I don't think that mkgmap is the right program for this. Maybe you can use different styles for that? On the other hand, I think the dead end check that is implemented now is incorrect because it doesn't report all "individual oneway roads that go nowhere" as documented. I did not find a simple way to fix that in the existing routine, but it can be done in the StyledConverter, because it is very similar to the check for unconnected roads implemented with r2530. Gerd