Add Lin_domain.stress_test

[jmid]

This PR adds Lin_domain.stress_{prop,test}
These are useful to test that parallel misuse does not cause any crashes.

For the testsuite we add stress tests of modules containing only negative parallel tests, that tend to find a counterexample fast, and hence not perform much parallel stress testing.

While experimenting with the branch, this mode has already been useful to help create a consistent reproducer of the Dynlink issue #307.

I tried cleaning up the git-history this time, so it should read relatively cleanly @shym 😉
Can you give it a brief look, e.g., just the first 4 commits?

[jmid]

CI summary:

• Cygwin trunk failed during configure [ocaml5-issue] configure flexlink test regression under Cygwin #439
• MinGW 5.2 timed out in the Dynlink stress test [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MinGW trunk timed out in Dynlink stress test [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• freebsd-amd64-5.0 failed with Failed: Internal error before running any tests
• freebsd-amd64-5.1 failed with Failed: Internal error before running any tests

The two MinGW failures nicely makes the case for this PR 🎉

Out of 35 workflows 5 failed with 3 genuine issues and 2 CI issues

[shym]

LGTM: the only thing I have to suggest would be to squash the first two commits together.
Otherwise, it is very nice to read, thank you!

[jmid]

CI summary for merge to main:

• Cygwin 5.2 found a counterexample in STM Out_channel test parallel [ocaml5-issue] Out_channel.flush can cause Syserror when used in parallel #444
• Cygwin trunk failed during configure [ocaml5-issue] configure flexlink test regression under Cygwin #439
• MinGW 5.2 timed out in Dynlink stresstest [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MinGW trunk timed out in Dynlink stresstest [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• freebsd-amd64-5.0 failed with Internal error
• freebsd-amd64-5.1 failed with Internal error

Out of 36 workflows 6 failed with 4 genuine issues and 2 internal multicoretests-ci failures