Potential bug in type_eq

We have

```ocaml
| EqDoNotFollowNull (* like EqStrict, but does not follow Null *)
```

But there's this check in the `TAnon,TAnon` case:

```ocaml
if f1.cf_kind <> f2.cf_kind && (param = EqStrict || param = EqCoreType || not (unify_kind f1.cf_kind f2.cf_kind)) then error [invalid_kind n f1.cf_kind f2.cf_kind];
```

This means that with `EqDoNotFollowNull` we allow field kind variance, which we probably shouldn't.

I would like to find a reproducible case instead of just blind-fixing this. There's on place in the inliner which uses `EqDoNotFollowNull`:

```ocaml
type_eq (if ctx.com.config.pf_static then EqDoNotFollowNull else EqStrict) etype tret;
```

This is to check if a `TCast` has to be inserted for the return value, so maybe it's possible to break this on HL which cares about `TCast` a lot.