Cleanup integration tests for stack repl

[ulidtko]

Overhaul of integration tests that exercise stack repl, gets them to run, and pass, on all matrix platforms (windows, too). Resolves #6170.

Glossing over various cleanups and refactorings (full details in commit messages), highlights of changes:

• eliminated threadDelay-driven "synchronization" in test logic;
• fixed missing EOF signalling on the child process' stdin at test finalization — resolving a -- TODO: Understand why the exit code is 1 despite running GHCi tests successfully;
• added logging of ghci's stderr on exceptions in test code — resolving a visibility issue / improving debuggability;
• byte-by-byte shuttling of stderr contents replaced with an FD redirect — yielding noticeable speed-up (in repl tests only);
• added -ignore-dot-ghci — its absence was causing failures in local runs on my machine.

Removed stubs that disable repl tests on windows, introduced in the wake of #6170 (PR #6174). They run again.

Test scenarios themselves kept intact; besides simplifications and cleanups, no logical change in asserts.

Tested via local runs, and CI spins in my fork. I got 5 consecutive runs 🟢 passing all 5 matrix jobs, on the same code (modulo rebases over updates in master). The repl tests are definitely running, and apparently stable.

Checklist:

• Get CI green 🟢
• Any changes that could be relevant to users have been recorded in ChangeLog.md.
  • not touching much beyond tests/, no need
• The documentation has been updated, if necessary

[ulidtko]

Hi @mpilgrem, this is ready for review, please take a look.

[mpilgrem]

You'll see from the 'linting' CI that Hlint has a suggestion. There are exceptions (see .hlint.yaml and .hlint-test.yaml) but, for the most part, the Stack project takes up what Hlint suggests.

The checking-in of the three stack.yaml.lock files are my past mistake. I've now deleted them from the master branch.

[mpilgrem]

@ulidtko, thanks. Looks good to me. You've not allowed edits from maintainers, otherwise I would have put through the small stuff/rebased directly. I've made some observations.

[ulidtko]

@mpilgrem comments addressed & rebased. Maintainer edits allowed, too.