Nearly all hardware support is kept in the kernel until and unless it bitrots to the point of unusability. I’ve had no issues with a 5.10-series kernel on my 2008 laptop, and I don’t expect any issues when I finally get around to upgrading it to 6.x (well, except the usual tedium of compiling a kernel on a machine that weak).
You’re unlikely to have issues unless an entire architecture loses support from your distro, and if you’re running x86_64, that isn’t going to happen for a long, long time. I’ve never been in a position where I couldn’t compile a new workable kernel for an existing system out of Gentoo’s repositories. The only time I’ve ever needed to put an upgrade aside for a few months involved a machine’s video card losing driver support from nvidia—I needed a few spare hours to make sure there were no issues while over to nouveau before I could install a new kernel.
Note that you can run an up-to-date userland on an older kernel, too, provided you make sensible software choices. Changes to the kernel are not supposed to break userspace—that’s meant to keep older software running on newer kernels, but it also works the other way around quite a bit of the time.
The difference isn’t all that noticeable, to be honest, or at least I’ve never found it so. If you’re using older hardware, you’re going to get an older “experience” anyway. The most user-visible kernel improvements tend to be improvements in hardware support, which is irrelevant if your hardware is already fully supported. However, I don’t do anything fancy with my machines—no full-disc encryption or the like. I usually don’t even need an initram to boot the system. So maybe you would notice something if your machines were more complicated.
(Note that the laptop I mentioned above started out with, um, a 3.x kernel? It gets a new one every year or so. The only kernel changes affecting it that were significant enough to draw my attention since 2008 were a fix in the support for the Broadcom wireless card it carries, and some changes to how hibernation works, which didn’t matter in the end because I basically never did try all that hard to get hibernation working on that machine.)
More recommendations mean more people using the hardware. More people using the hardware means more testing. More testing means more people learning and documenting how to fix problems. So in that sense, statements like that actually do become true over time regardless of their truth values at the beginning.
You’re using software that’s being continuously developed by people for whom stability of the UI is not a priority. Pointless UI churn is normal. Half-assed solutions kept beyond their best-before date are normal. Windows does this crap too. At least with Linux you have a choice of which issues you’re going to tolerate (or you can pick a DE where UI stability is a priority for the development team).
Some native distro formats are unlikely to ever be supported by services of this type. For instance, neither of the two services you list in your opening post will generate Gentoo ebuilds, most likely because the process is fundamentally different: an ebuild is a set of instructions for the package manager, not a prepacked binary.
Linux mostly follows POSIX standards, even though it’s never been certified as compliant, so much code targeting POSIX systems runs on Linux too. In other words, it didn’t establish any standards so much as adopt one that already existed.
There is no POSIX standard for package managers, however.