Phasing out EOL distributives

[develop7]

@develop7 you're right. I removed all unrelated CI changes from this PR.
I commented out the buster stuff out of frustration because I didn't understand where the failing jobs are coming from.
(It was because I was missing the exclusions for 9.12 which doesn't have buster/bullseye variants.)

But in general are you in favor of both removing buster from CI as well as removing all buster dockerfiles from the repo in separate PR?
My arguments for removal: 1. don't waste CI resources building something that's past the EOL. 2. make the repo easier to navigate 3. if anyone wants those buster dockerfiles they can find them in git history.

Originally posted by @jhrcek in #151 (comment)

[develop7]

I think distros long in EOL definitely deserve phasing out, but stopping them from building altogether? Nah. After all we still have deb9 bindists even for 9.12.

I'd suggest to make buster jobs "allowed to fail" instead, or extract them to separate workflow even.

[develop7]

Resources consumption-wise, I don't think it is a problem, unless GitHub would say so. Longer CI times, OTOH, increase the feedback loop, increase friction and harm productivity, especially on a side volunteer-powered project like that.

[develop7]

@jhrcek what do you think?

[jhrcek]

I don't have a strong opinion at this point.
Marking the "*-buster" jobs as not "allowed to fail" would makes sense to me at the minimum.

[jhrcek]

Related thing we should update is this note in dockerhub description (which comes from this repo):

Note: Currently stable Debian is version 11 bullseye, however it is not yet supported by Haskell tooling. Until that time the default will remain Debian 10 buster. We have dropped support for Debian 9 stretch.