The main problem (from my point of view) of python data science ecosystem is a complete lack of standardization on anything.
You have ten different libraries that try to behave like 4 other languages and the only point of standardization is that there is usually something like .to_numpy() function. This means that most of the time I was not solving any specific problem related to data processing, but just converting data from a format one library would understand to something another library would understand.
In Julia (a language with it's own problems, of course) things mostly just work. The library for calculating uncertainties interacts well with a library handling units and all this works fine with the piloting library, or libraries solving differential equations etc. In python, this required quite a lot of boilerplate.
Nobody has mentioned array-api (and data-apis more generally), which is trying to standardize the way people interact with arrays across the Python ecosystem.
Sounds like a great idea, but difficult to achieve. The announcement blog post was almost 5 years ago, do you know maybe what the impact of this project has been in practice?
R with its 4(?) class systems enters the chat.
It's at least 5 at this point.
In defense of R the class systems do have different characteristics and they are not deeply embedded in the language or anything.