Expand STM Semaphore tests

[jmid]

While trying out a run of multicoretests under TSan, I spotted some limitations of our STM Semaphore tests. For one, we test Semaphore.Counting.get_value in parallel despite it being documented as racy.

This PR

• includes a parallel variant of the existing STM Semaphore.Counting test that excludes get_value
• adds a corresponding Semaphore.Binary STM test for completeness
• renames files accordingly

Because of the precondition in the Semaphore.Binary STM test, it needs to generate extra inputs. By default, qcheck-stm tries to generate up to 3x as many inputs as needed which is not quite enough to hit 500:

$ dune exec src/semaphore/stm_tests_binary.exe -- -v
random seed: 338133583
generated error fail pass / total     time test name
[✓]  500    0    0  500 /  500     0.1s STM Semaphore.Binary test sequential
[✓] 1500    0    0  332 /  500     2.1s STM Semaphore.Binary test parallel
================================================================================
success (ran 2 tests)

The test is still relatively fast with a count of 500 compared to the STM Semaphore.Counting test which has a count of 200:

$ dune exec src/semaphore/stm_tests.exe -- -v
random seed: 72247306
generated error fail pass / total     time test name
[✓]  200    0    0  200 /  200     0.0s STM Semaphore.Counting test sequential
[✓]  455    0    0  200 /  200     2.1s STM Semaphore.Counting test parallel (w/ get_value)
[✓]  600    0    0  179 /  200     1.7s STM Semaphore.Counting test parallel (w/o get_value)
================================================================================
success (ran 3 tests)

As such I'm leaning towards declaring good enough...

[jmid]

CI summary:

• 32bit 5.1 found an unexpected counterexample in Lin.Internal CList int test with Thread #358
• Linux 5.1 debug triggered an assertion failure in thread_joingraph #353
• freebsd-amd64-5.1 failed to trigger a STM Sys test parallel counterexample #410

Out of 63 workflows 3 failed with 1 genuine error and 2 false positives

[jmid]

CI summary for merge to main:

• Cygwin trunk timed out during Dynlink test #307
• MinGW bytecode 5.1 timed out during threadomain #203
• linux-s390x-5.1 timed out on s390x-worker-01 #421
• macos-arm64-5.1 triggered an unexpected Sys counterexample #359
• freebsd-amd64-5.0 failed with Internal error

Out of 38 workflows 5 failed with 2 genuine issues, 2 CI issue, and 1 test suite false alarm.