You could argue that we should adopt the same testing strategy of mission critical pieces of logics where 1 LOC has 1+ LOCs of test code, aim for 100% coverage, ... . But then your windows license would cost 1000s of $.
And about your final remark. What you call real engineering has to cope with well defined requirements and the context is well known. You know beforehand the maximum weight a bridge should be able to support. You know the configuration and properties of the soil where the bridge will be built, and you have good estimates of 1 in 100 year extreme events magnitude.
Software doesn't have that luxury. If we build a web server to handle 1k qps it still is somewhat likely that said server might face spikes of 10k qps. Try to do that with the bridge mentioned above. Forget failing gracefully. Additionally users have few ways to use the bridge. You either cross it by car / bus / on foot. The same web server might face someone trying to send a payload of GBs where you would expect few KBs. That's at least partly why code is messy. The space of possibilities is much greater and we somehow need to write software that still works.
