In addition to Chris' point, the existence of "short" dead ends is one of the reasons that we use multiple connectivity measures. So we measure the length of deadends as well as the fraction of edges that are deadends. And these short dead ends will also not affect our measures of circuity.
We were reluctant to use a cutoff like 20m, because the correct cutoff will vary by country. (In the US, there are some very long driveways!).
However, most driveways are (or should be) tagged as "service" roads in OSM. Like pedestrians and bicycle paths, we include service roads in calculating the connectivity of an area, but we do not treat them as destinations. The Dresden example you raise is tagged as "residential road," not "service."