You’d think certain patterns could be probably safe and the interpreter could take the initiative.
Is there a term for this concept?
It is another one of those things that many programmers ask "Why hasn't this been tried?" and the answer is, it has, many times over. It just failed so hard and so fast you've never heard of the results. Toy problems speed up by less than you'd hope and real code gets basically nothing. Your intuition says your code is full of parallelization opportunities; your intuition turns out to be wrong. A subset of the general problem that even very experienced developers still never get great at knowing how code will perform without simply runninga profiler.
It has failed hard and fast in languages much friendlier to the process than Python. Proving something is truly parallel-safe in Fortran is hard enough. Proving it in Python is effectively impossible, you just don't know when something is going to dynamically dynamic the dynamics on you.
I'm going to use that!