I disagree. The scientific libraries were just one of many niches, and an awkward one at that. One could equally say "it started with xml libraries" or "it started with file-handling libraries" or "it started with http libraries" -- all of which were in the Standard Library very early, unlike the horrible-to-build numpy/scipy. All of these made the language popular initially across a number of different crowds. Numpy/scipy reached traction relatively late when Python was already well-established in niches like sysadmin, web, education, 3D, and many others. By 2001 we already had multiple web frameworks, Zope, even WSGI...
It is occasionally annoying how this or that crowd tries to appropriate Python's success, hence flattening its purposes and aims.
In 2000 I joined a new startup and was shown Python 1.5.2 by the startup's chief architect/scientists. He'd come from Infoseek where he'd used it there to help build parts of the company.
Now, I loved Perl. I was one of those annoying geeks with "RSA in 4 lines of Perl" T-shirt. I'd write JAPH programs for fun. But I appreciated how much "smaller" Python was than Perl. I was able to learn pretty much all of its rules in an hour? I didn't have to worry about all the crazy ways Perl programs could express themselves. (Is this scalar context or array context?[^1] Is this script using my or local? What the heck does this crazy line I didn't write do again? Oh wait, I wrote that line!)
So anyway, this startup was using Python and now so was I.
That startup is long gone, but I'm still using Python daily in my career.
And I've never used numpy.
Perl was/is great, but it's just too quirky to have the broad appeal of Python.
[^1]: "It's all about context": https://archive.ph/IB2kR