I thought the Sapir-Whorf hypothesis only predicted that the difficulty in understanding certain things would be language dependent, not that it would be impossible per se. To make an apt analogy, some languages force you to declare a whole bunch of factory methods and boilerplate bloat before you can express a program that prints "hello world", others simply let you write print("hello world").
The strong version basically says that a person's worldview is just about completely determined by language.
This was pretty much tossed in the trash bin, partly due to an interesting study into how a language's lexical entries for colors influences perceptions on color closeness and categorization.
Instead of the strong version there's a reasonable consensus that language influenced things but does not wholely determine them.
An interesting example is that speakers of tonal languages are more likely to exhibit perfect pitch.
Source: my increasingly hazy recollections from a post-graduate comp ling program.
It's a bit tricky to say what the "Sapir-Whorf hypothesis" predicts, since apparently neither Sapir nor Whorf exactly formulated a clearly stated hypothesis on this topic. This means people write about "strong" or "weak" versions of it.
I think, to use your analogy, any language that lets you write new libraries which can be imported, will tend to become pretty decent at anything which people programming in that language do a lot. Whatever problems there are in the language itself, tend to become ameliorated (though probably not entirely eliminated) by focused work, for example spinning up a neural net or scraping a website gets much easier once a lot of people have done it in your language of choice, and they have released a library that they use to do it.
So, a language may not be good for speaking about a topic which the speakers of that language don't have much experience with, but if they come to have much experience with it, the language will quickly evolve to get better at it.
