There's an interesting point here. I suspect we would both agree that "driving has become safer in Colorado" because more people are choosing to be stoned at home, instead of drunk on the roads. However, I think there are still a lot of people stoned on the roads, and that driving would become even more safer if we could deter that through reliable detection. But it's possible that would be a wash or have a negative result due to re-tipping the scales in favor of being drunk and potentially driving. Obviously, that hasn't been tested, but I think it's worth shooting for an even better outcome, rather than just avoiding an overall bad one.
You have already pointed out why that might not work in practice, plus you have not proposed a way to prevent law enforcement from imposing de facto prohibition or creating other self-serving distortions as law enforcement often does.