Your question takes a very narrow and untrue view of what computer science is. Computer science doesn’t “assume a static problem” — there is an entire subfield studying Online Algorithms, which take input problems piece-by-piece, as just one example.
Computer science is the system of methods for analyzing computing problems (i.e., making efficient use of computing time and computing memory). If you have (correctly) worked out a better way to solve a problem, then you are doing Computer Science. In general, it makes no such “static assumptions” or “ignores in-memory pointer solutions”. You’ll find that nearly any problem has been studied from nearly any angle, by someone in computer science.
When you describe “a system of pointers and how to store them in memory”, what are you doing but describing an algorithm? An algorithm is a description of executable steps — if you have devices a system of pointer-bookkeeping that solves the problem, you have devised an algorithm. If that algorithm doesn’t find the “optimal” solution, it’s still an algorithm that may be analyzeable by standard methods in Computer Science (e.g., “average”/worst-case runtime; or approximation bounds)
It’s true that many problems faced in the “real world” are not addressable solely by the algorithms and data-structures taught in the first few years of a computer science education, and also that many real word problems do not require optimal solutions (not that computer science only studies optimal solutions — it doesn’t, optimizations and approximations are very intently studied).
However, the point of teaching the “standard” algorithms and data-structures isn’t only to give a complete toolkit, but to provide a breadth of examples that introduce many different techniques:
- Algorithms for solving NP-complete problems like the Traveling Salesman Problem can introduce the concept of heuristics (which may or may not be “complete”) and approximations
- Data-structures like doubling-arrays can introduce the concept of amortized analysis
- Data-structures like balanced trees prove out the concept of worst case analysis, showing that naive implementations can lead to pathological runtimes, and showing that there are bounds to how much you can improve in general (e.g., you can do insertion & lookup better than $Theta(n)$ and no better than $Theta(log(n))$)