This looks good to me, modulo the nitpick that the order of the
val declarations in the .mli could respect the order or the
corresponding let declarations in the .ml.

Thanks for the fast PR.

I'm still wondering what to do for ocamlbrowser on ocaml 5.2.
This PR will eventually allow to write a 5.3 version of ocamlbrowser, but for now the only solution still seems to be to include a copy of cmt2annot.ml. The license of ocamlbrowser being the same as ocaml, is this OK ? (They were originally distributed together.)
(I just realized that there is no proper license included in the labltk distribution; it is only indicated in the opam package, but I will fix that.)

The license of ocamlbrowser being the same as ocaml, is this OK ?

Yes, this is OK. The people who wrote and then made substantial changes to this file will become contributors of ocamlbrowser, which is fine.

Note: I think that it's not necessarily a bad thing that you adopt this part of the logic and have the opportunity to adopt/refine/simplify it based on the needs of ocamlbrowser. Maybe you will find in the end that this is better than hooking into this specific compiler-libs module, time will tell.

Would it help if we'd backport this fix to 5.2, assumingthere will be a 5.2.1 release at some point?

I don't think it is necessary. For this release, compatible with 5.2, I will include cmt2annot.ml in the labltk distribution.
For the next one, based on 5.3, I can then remove it.