Kidding aside, someone has to build those tested, stable libraries that handle those problem (or even untested bleeding-edge if you are breaking new ground).
Due to the circumstances, normalization wasn't an efficient option. I ended up throwing together a barebones tree with a 5-line DFS implementation to traverse it. It handled inserts, updates and deletions (for my use-case) in linear time.
The details aren't so important as the fact that adding a dependency would have been overkill for my needs. This isn't to say that efficient graph implementation libraries should not exist or be used, but I was able to produce this code faster by having that basic CS knowledge.
Context and knowing the right tool for the job is important indeed.
At the level of programming that the grandparent is talking about, I'd accept the judgement of the programmer working on it as to the appropriate solution.
Two big areas that come to mind are simulation/mathematical modelling, where you're often crunching data in ways that aren't just textbook examples, and embedded systems, where you often have resource constraints that make efficiency more important.
This doesn't just mean modelling weather systems on supercomputers or writing the control software for cars, though. For example, consider user interfaces. We are increasingly looking for more intuitive input methods using techniques like natural language processing, speech recognition, handwriting recognition, and gesture-based UIs. We are looking for more intuitive output methods, such as integrating additional data with real world imagery like maps or the view through a 3D head set or camera. We are looking for systems that learn patterns in their users' behaviour and adapt to provide more likely options more quickly next time.
You won't see much of this if you're just writing simple form-based web front-ends for CRUD applications. A lot of real world software is like that, and it gets a lot of useful work done, but it's mostly pretty mundane, join-the-dots work as far as the programming goes. However, there are plenty of interesting problems out there and we could directly improve the user's experience in new and helpful ways if we could solve them, and much of that work involve developing data structures and algorithms far beyond anything you'd find in an introductory textbook.