When it comes to Android 2.0-or-so phones, this is already possible in many cases. Just grab an old CyanogenMod image from the archived version of the site, grab kernel sources from the Lineage OS git repository, and start hacking on support for that phone as part of the postmarketOS project. The real problem is that every single one of those phone models runs on bespoke hardware, there's no standardization like you would find on the PC platform. It's a lot of effort for an unclear gain. A startup is not going to do this, especially when those phones are irredeemably insecure against sophisticated attackers, due to proprietary implementations in the baseband and WiFi/BT radio components.
Unfortunately these phones are incredibly under powered, they come from a time when arguments for bizarre custom OSes like Android actually still made some sense (not much IMO and after the generation that used android 4 it was really all about politics, economics, and control more than a technical argument.) The outdated radios are another issue; even WiFi evolves to the point where old enough devices stop being able to join some networks but the cellular interfaces are particularly bad.
The whole bespoke hardware thing is only an issue when you're updating the kernel (I've heard device tree debugging has gotten much better so maybe this is an opportunity to try and fix that heh), as long as the arm chip is new enough to use the modern calling conventions all of the distribution binaries should just work (I guess some of the earliest phones can't do this like the really ancient freescale SOCs etc. For those you're just stuck building everything yourself but with modern user spaces this really isn't as bad as it sounds.)
Not much, IMO. As far as compute goes, those old phones were approximately as powerful as a Raspberry Pi 1 board, with plenty of RAM as well (512MB at a bare minimum). And a conventional Linux OS can run just fine on those specs, of course.
(GPU acceleration is actually more important to have, because you can't have a modern mobile UX without it. Even Purism is very possibly going to have issues on that front, BTW. GNOME-Mobile/Phosh is a nice prototype effort, but I do not regard it as genuinely usable just yet.)
I seem to remember a few phones with 128 meg, but looking at a few at GSM Arena they all have 192MB or more. I have my old Android dev phone (upgraded all the way to Cupcake!), and that has 192MB RAM. I wonder if that still starts.
The problem is that a lot of apps start misbehaving. Firefox and Libreoffice come to mind. Firefox is really impossible to replace too (IMO) I haven’t been able to log in to gmail from a gtk WebKit browser in years and medium doesn’t even work in them (it of course works in elinks.) You can run Firefox in <1GB of ram but it won’t be nice and it won’t run long.
I absolutely disagree with you about the GPU. As long as you’re not running gnome software graphics are plenty fast.
I couldn't agree more. Unfortunately not off the shelf hardware is expensive, software support for it (device drivers) is even more expensive if not plain impossible to obtain from manufacturers in the form of open documentation, and all off the shelf phones prevent any open operating system from running on them by design, therefore making a new platform from scratch is the only viable albeit very expensive solution.
I wish Purism, Pine64 and others best luck, however. In the meantime I can only dream of laws that would force hardware manufacturers to release technical data about hardware after either X years or once they declare it obsolete or some time after they stopped selling/supporting it. Just think of the huge decrease in pollution when a startup could take all those Android 2.0+ phones, all those junk iPhone clones from 5-7 years ago, and convert them say into home controllers, portable music players etc. Old phones became junk over time because the manufacturers programmed them to become as such by refusing any kind of repurposing.