Initial STM precond fix, add Domainslib.Chan tests

[jmid]

This PR provides an initial implementation STM precond support for parallel mode using a generate-and-filter approach.
Following Claessen-al:ICFP09 we keep only cmd list triples where all interleavings satisfy all preconditions.

The precond implementation is enough to get the second part running:
This part tests the Domainslib.Chan module. This requires precond to avoid

• recv blocking on an empty channel
• send blocking on a full channel

For the Domainslib.Chan tests, this means that we need to generate around twice as many triples as the test count.
I've also checked that the run time of the other STM tests is not affected noticably. For reference we now have the following STM tests:

src/atomic/atomic_test.ml
src/buffer/buffer_stm_test.ml           ;; expected to fail
src/domainslib/chan_tests.ml
src/lazy/lazy_stm_test.ml               ;; expected to fail
src/lockfree/ws_deque_test.ml
src/neg_tests/conclist_test.ml          ;; expected to fail
src/neg_tests/ref_test.ml               ;; expected to fail

Ping @n-osborne 👀

[n-osborne]

As the prefix is sequential, it should be valid by construction (?). So, do we really need to check it here or could we just compute the spawn_state?

[jmid]

Good point! 👍
From a generation perspective I completely agree.
From a shrinking perspective, I don't think we can be sure that the prefix continues to be valid precond-wise, e.g., if the cmd list shrinker removes cmds from the sequential prefix...

Such shrinking reminds me that it would be good to actually test that this combination works - a negative test (triggering shrinking) with a non-constant precond....

[jmid]

Rerunning jobs as trunk has now been patched for the MacOS segfault.

[jmid]

Rebased on top of #63 - also bumping max_gen to 3 * count...

[jmid]

Ran this manually on the MacMini also - 5 times.

• 3 runs were all green
• one run triggered the Buffer failure (first time on MacOS I think)
• another run failed to trigger the expected 2 failures in src/lazy/lazy_stm_test.ml

The latter only runs 100 iterations though - so bumping this count to 200...

[jmid]

Runs fine on manual MacOS tests - merging