Avoid using _Bool in public headers for the sake of C++ compatibility

[gadmm]

Headers must be compatible with C++ so they cannot refer to _Bool. We replace _Bool for Caml_state->action_pending with intnat where only the least-significant byte is taken into account.

Fixes: #12772. See discussion there regarding the choice for s390x.

Precheck passes: https://ci.inria.fr/ocaml/job/precheck/915/

[kit-ty-kate]

I believe a similar change needs to happen in the 5.1 branch as https://github.com/ocaml/ocaml/pull/12734/files#diff-92e5994819665dce8d9ae0342af2dbb1ff3cd11666eb080873f54c2cc04a5ca3R138 introduced a use of _Bool as well.

[gadmm]

This other occurrence is under a CAML_INTERNALS guard so I am afraid that this part is not covered by warranty :)

I checked and the current modification is the only occurrence of _Bool in need of removal.

[kayceesrk]

LGTM.

Do we have tests in the testsuite that fail if the RET_FROM_C_CALL macro does not work as expected?

[xavierleroy]

Even if the remaining _Bool in caml/*.h is protected by CAML_INTERNALS, you're welcome to replace it with int for consistency.

[xavierleroy]

Looks good to me.

[gadmm]

We test that the poll happens when it should. We do not test that no poll happens when it should not (and it would ask some effort to devise one).

I'm a proponent of using bool inside the runtime, which sometimes leads to using _Bool inside internal headers. All the more as the situation will resolve itself in the long term with C23.

A test that would be nice to have in the CI is that C++ compilers understand public OCaml headers.