Run s2n tests for only one GHC version

[atomb]

This may not fix the timeout issues we're seeing, but at least it'll reduce the amount of work we're doing, and reduce the number of tests that might timeout.

Also fixes several other CI issues discovered along the way.

[robdockins]

I don't know how easy/hard this is to do, but it would be nice if the s2n tests only depended on the particular build that they need in order to run. As it is now, they wait for all builds to successfully complete before beginning. In particular, if one of the builds fails because of some configuration situation, the proof tests don't run at all.

For example, the Windows 8.8.4 build currently seems like it might stall until it is canceled, and the proof tests will never run, preventing merging.

[atomb]

I entirely agree with that, but I unfortunately also don't know how hard it is to do. At the very least, it'll require learning some more about how GitHub Actions work.

[kquick]

Would it be better to make this if: runner.os == 'Linux' so that it's immunized against future ubuntu release updates?

The concern here is that the matrix entries might be updated without checking the rest of the file for matches and suddenly this step silently stops being run.

[pnwamk]

N.B., #1429 includes the fix from b66f637 in this PR (along with some other server-related CI fixes)

[atomb]

At this point, this PR is actually all about updating GHC and Z3 versions, which may not actually be something we want to do right now.

[atomb]

The important parts of this have been incorporated into other PRs, and the remainder has been re-done in better ways.