Update supported GHC list

[parsonsmatt]

This PR drops GHC < 7.10 support and adds CI for GHC 9.0 and 9.2

According to the State of Haskell 2021 Survey, fewer than 1% of respondents were using GHC < 8.2.

GHC 7.10 was released seven years ago, which is a pretty huge maintenance window.

It's difficult to even install or build things with GHCs that old. I was able to get the repo building with stack and --resolver lts-6, but prior resolvers had problems with Cabal and needed older stack versions. ghcup doesn't provide anything older than 7.10.3, and I wasn't even able to get that to build the library on my machine.

[parsonsmatt]

CI was failing because no build boxes were available for this for over a day. Trying this to see if it'll work

[parsonsmatt]

@sol This is a pretty big change in terms of what is being dropped, but I would be really surprised if anyone is impacted by this. There are, after all, lots of hspec versions that claim to support older GHCs. This gets CI passing and allows other things to get merged in more easily. Any objections?

[sol]

@parsonsmatt I think if we use ubuntu-18.04 on CI it should be easy to keep support for older versions of GHC (as we do for hspec).

[parsonsmatt]

Indeed, getting the ubuntu-18.04 box in CI got everything passing back down to GHC 7.4.

I am curious what the point of supporting GHCs so old is. Is this a firm commitment (eg we won't incorporate features that will break for old GHC / we'll use CPP to preserve old GHC compatibility), or more of a "well we'll keep it working as long as we can, and won't break it until we know we have to" sort of thing?

[sol]

@parsonsmatt not a big thing, but I guess we should remove the latest from the comment. Other than that it's good to go.

[sol]

@parsonsmatt please squash the commits and rebase.

[sol]

I am curious what the point of supporting GHCs so old is. Is this a firm commitment (eg we won't incorporate features that will break for old GHC / we'll use CPP to preserve old GHC compatibility), or more of a "well we'll keep it working as long as we can, and won't break it until we know we have to" sort of thing?

I'm willing to discuss dropping support for older GHCs at the point it provides tangible user value.

[sol]

Closing in favor of #59.

[sol]

Rational: The track record of haskell/actions/setup is not that great (e.g. haskell/actions#80 - silently used the wrong version of GHC). Also, the repo structure does not allow for publishing to GitHub marketplace. I'll do a more thorough write up in the README of hspec/setup-haskell when I get to it.