Promote val_modalities to Stable

[riaqn]

This PR promotes modalities on values descriptions and module declarations to Stable.

There are no longer behaviors controlled by mode_alpha, but maybe it's fine to keep it for future convenience.

I confirm that our internal code base builds without changes, with this PR.

[tdelvecchio-jsc]

Looks good.

[tdelvecchio-jsc]

I'm currently running into an issue where

(* x.ml, with no mli *)
type t = T of unit

let t = T ()
let u = lazy t

(* y.ml *)
let _ @ uncontended = X.u

(* Error: This value is contended but expected to be uncontended. *)

I hypothesize (but have not checked) that this PR introduces this bug. Can you check?

[riaqn]

This is #2967 missing a mode crossing spot. Will fix.

Since we will push u to stongest modality portable, that makes the implicit lazy closure fun () -> t portable, which makes t only available as contended in the closure, which makes the returns of the closure contended and thus the whole lazy is at contended.

[riaqn]

Well I just realized that the said mode crossing is already there for general variable lookup. What's missing is that the kind system don't know type t crosses contention. You can either try #3284, or maybe unsafely specify that it crosses contention?

Do you encounter similiar issues where such type t contains mutable fields? That might be more tricky to think about.

[goldfirere]

I think this should also update the documentation, now that we're putting the feature in stable.

[riaqn]

@goldfirere Good point. Then we need documentation for portability and contention first, because val_modalities is mostly for them.

Also, we agreed to revise the new mode syntax so that @ is for modes and @@ for modalities. We need to do that revision before writing any documents.