More tests for picos.sync

[polytypic]

I may have observed the picos.sync tests potentially dead/livelocking at least on (32-bit) OCaml 4.14 on CI. This might indicate a bug in the picos.sync library, a bug in the test, a bug in (32-bit) OCaml 4.14 (I don't recall seeing the test not completing on other OCaml versions, but I might have simply missed that), or it might be a completely unrelated thing (test machine being slow for some other reason). At any rate, this needs to be investigated further and the correctness of the picos.sync library implementation ensured.

Observations:

• debian-12-4.14_arm32_opam-2.1 (not completed after 24+ minutes, completed very quickly after cancel+rebuild)
• If thread-local-storage is (for some reason) not installed, it was possible, before When threads.posix exists we really need thread-local-storage #110, to build a non-working set of libraries where the mutex cancelation test and benchmarks did not terminate. This shouldn't really be the case with the observed non-completion.
• Tried running the picos_sync test repeatedly in parallel (dozen or so) with OCaml 4.14.2 on macOS with M1. Did not get any lockups within a few hours.
• debian-12-4.14_arm64_opam-2.1(not completed in an hour)
• debian-12-4.14_opam-2.2 (seemed to be stuck in the cancelation test)

[polytypic]

It might be that the issue was related to the cancelation test spawning fibers, which translate to systhreads on OCaml 4. PR #230 changes the tests to not spawn fibers. Time will tell whether this eliminates the hangs on OCaml 4.

Addition: There was a test run where the cancelation test did not seem to complete on 4.14 arm64. Not spawning lots of systhreads seems to have made the failures less common.

[polytypic]

@edwintorok mentioned about the pthread_cond_wait bug, which might be the cause of the issues.