Put Unix.SO_REUSEPORT at the end of data variant

[dra27]

In #9869 (comment) I noted that the new SO_REUSEPORT should ideally have gone at the end of the variant for vague C compatibility reasons.

However, there's another possibly stronger reason and, given that we're still at alpha, I propose we do actually change it. Here's an error message from Lwt, prior to ocsigen/lwt#804 being merged:

File "src/unix/lwt_unix.cppo.mli", lines 1020-1030, characters 0-13:
Error: This variant or record definition does not match that of type
         Unix.socket_bool_option
       Constructors number 4 have different names, SO_REUSEPORT and SO_KEEPALIVE.

which looks much more serious and less obvious than, with this PR:

File "src/unix/lwt_unix.cppo.mli", lines 1020-1030, characters 0-13:
Error: This variant or record definition does not match that of type
         Unix.socket_bool_option
       The constructor SO_REUSEPORT is only present in the original definition.

(cc @kit-ty-kate)

[dra27]

Another option, of course, would be to increase the sophistication of the error message even further (so spot that one set of constructors is a subset of the other)

[Octachron]

The error message could be amended, but the two messages seems quite comparable to me. Moreover, fixing the error highlighted in the first message will yield the second error message.

[gasche]

The change looks reasonable (and, in particular, correct) to me, and I think that adding new fields at the end is indeed a good practice.

[dra27]

@Octachron - the present message is of course technically correct, but it doesn't lead towards the likely correct remedy (it's reminding me of this joke!).

[Octachron]

The current error message does lead towards the correct remedy even if the user follow it blindly? It is the number of steps required that is suboptimal. And I expect most people to simply resynchronize the re-exported definition rather than trying to fix this error by hand.
Note that I am neutral to the change, this particular motivation just feels dubious to me.

[gasche]

I am happy with the change. Should we have it in 4.12? (I think yes.)

[dra27]

This needs to be 4.12 or not at all - I think it would be bad to have 4.12 with SO_REUSEPORT = 3 and 4.13+ with SO_REUSEPORT = 9

[kit-ty-kate]

Even though this is breaking change with previous alphas I'm all for this PR.

That I can remember, only batteries and lwt will need to change again after this.
cc @raphael-proust would it be ok to squeeze that before the release?
cc @UnixJunkie would it be ok to release batteries 3.2.1 with this change?

[UnixJunkie]

@kit-ty-kate we can release a bugfixed batteries version if needed. No problem.

[kit-ty-kate]

Ported to lwt (ocsigen/lwt#826), batteries (ocaml-batteries-team/batteries-included#992) and core (janestreet/core#142)