> There was a small period of time between the 90s and the 2010s where you could grab almost every 386 OS and have your hardware mostly decently run for it
And prior to that, you could grab every OS running on IBM clones and not have to worry about graphics drivers at all, because graphics acceleration wasn't a thing. The era you refer to had already introduced software contingency on hardware within x86. This disparity was further compounded in the mid-2010s as GPUs exploded in complexity and their drivers screamed into tens of millions of lines of code, eclipsing kernels themselves. This is not distinguishable from the introduction of graphics drivers in any generalized manner. They were driven by the same process.
An important thing I want to point out as well; you're doing a lot of heavy lifting by limiting the pool to x86 computers, which is already giving up and admitting to a very strong restriction of hardware choice. Don't take that as pedantry, it's a very well hidden assumption that you've accidentally overlooked, or in the case that you think it's irrelevant, I'm letting you know that I don't consider it irrelevant in the slightest. When I think of computers, I'm not just thinking of x86 PCs. In the 90s I'm thinking of SGI workstations, Acorns, Amiga, Macs. I'm thinking of mainframes and supercomputers and everything else.
> This is the wrong logic, because operating systems become esoteric since they can't support the hardware
On the contrary, I assure you that this logic rests on faulty premises. As a general principle it's clearly false since most operating systems (which are long forgotten) predate it by decades, and in the specific context of Linux winning over FreeBSD, it's still not applicable as that happened smack dab in this era you describe.
> You may _think_ you have no reason to be sad if you're a user of Windows or Linux, but you have plenty anyway.
I'm a user of Linux, FreeBSD and 9Front. I just don't (and never have) bought hardware at random. You can reason your way into sadness any which way, but rationalization isn't always meaningfully justified. I just don't find it sad that my second desktop can't have an RX 9000 whatever in it. Where's the cut off line for that? Why not be sad that I can't jam a Fujitsu ARM processor into a PCIE slot as another type of satellite processor? The incompatibility is of the same effect, but I don't see you lamenting or even considering the latter, as though mounting a processor to a PCB is somehow fundamentally less possible than writing a modern graphics driver.