FTP was older and had more widespread client support in the 90s, so you were more likely to have at least a basic client preinstalled. That's obviously moot by now since just about everything ships with a web browser and/or something like curl.
FTP has a standard way to list directories. Using HTTP that way would require you to either have a well-known index file or parse HTML looking for links which don't return text/html responses.
The downside is that FTP still has issues with firewalls – I had to troubleshoot that earlier this month, actually – and is another service to maintain if you are already running an HTTP server.
In the case of either single-file downloads or something like a Linux package manager, the URLs are well known so directory listings are irrelevant. HTTP has a number of good options for CDNs and caching, so if you care about performance or reliability that's a turn-key service.
In the cases where directory listings were more valuable, people usually wanted a richer UI than just a file listing, too, and there are tons of options for that in the web world.
In the old days, we had decent clients for http but they were all manual operated GUIs, and the decent clients for FTP were all CLI and/or extremely automatable. Also in the old days it was unusual to have a machine with a GUI maybe 30 years ago.
Also there was a day before dependency resolving automatic downloading GPG verifying distribution clients. I was there... Say you needed to roll back (or security upgrade) some software, you'd quite possibly go to a FTP site thats trusted, download some tar.gz or tar.z or shar archive that was trusted, make, make install, plus or minus some configuration of course. You can't do that via http in 1990 while telnet'd into a server, the http infrastructure and commands and servers hadn't been invented yet.
Its also important to point out that at least some of us were fooling around with unix and FTP (and FTP to non-unix OS, I vaguely remember some TOPS20 server in the late 80s early 90s... simtel20?) before http ever existed. So naturally there were a lot of tools and processes and experience getting things done with FTP when HTTP arrived.
Decent FTP client CLIs were extremely advanced. Not exactly "wget someurl" and hope for the best LOL. Automatic login with differing accounts per site, text mode graphs and stats as downloads commence, tab completion, local help command functionality, multi-connection support... Mirrored and often lead the developments in modem/BBS download functionality (think like Zmodem on Telix not kermit)
Its sort of like those "how could you have thought using telnet was a good idea compared to ssh?" Well, I had 15 years of unix experience before OpenSSH was deployable, so we had a lot of telnet experience...
It's hardly sudden! The last time I encountered anyone using FTP by choice was around the turn of the millennium. Since then, it's always been supported reluctantly and under protest, because some integration partner critically important to a profit center couldn't or wouldn't support anything less terrible.
All the reasons listed in the article. It's also insecure. It's Hell with firewalls because of the way it negotiates what ports to use, instead of using a standard port for data and control.
I'm a nostalgic kind of guy, but FTP is terrible through firewalls and nowadays, bittorrent or http are just as well.