Rebase fixLinkDeps onto solver branch

[grayjay]

All but the last two commits are from #3359.

See #3357. I rebased https://github.com/edsko/cabal/tree/pr/FixLinkDeps, except for the last commit, which handles a different issue. I also removed my commits, because they weren't needed.

/cc @edsko

[edsko]

@grayjay I'm somewhat confused. What is the status on #3327 now? Do we have a solution? Does the solver branch now have my fix for the incorrect link groups (the unqualifation-requalification thing) and your original solution to #3327? And the discussion at the end of #3357 (comment) we still don't have a good answer for? Is that correct?

[grayjay]

@edsko This PR only contains your solution to #3268, which uses unqualifyDeps. I wasn't sure what to do about #3327. I think that my PR is incomplete because it doesn't contain GoalReasonChains, and we haven't decided what to do with lgBlame and lgConflictSet.

[edsko]

Yes, I think the answer to that isn't clear yet, but it would be good if we have have the test cases you have for #3327 available on the solver branch (where we expect them to fail right now).

[grayjay]

I cherry-picked that test case, but I think it is currently passing because of a different fix.

[edsko]

Aha, I see. That explains my confusion :) That's a little unfortunate.

[grayjay]

Yeah, there is a lot of overlap between conflict sets. I'll try to come up with a more specific test.

[grayjay]

I added a test in #3363.