Improve Inlay Hints Handling with [@merlin.hide]

[mlasson]

Description

Previously, inlay hints logic relied on ghost locations to avoid annotating generated code. But it lead to some inconsistencies because ghost locations are used in non-generated code, for instance:

let f = fun x -> x ^ ""
let g _ = fun x -> x ^ ""

In this case, an inlay hint appears for x in f, but no hint is shown for g.

In VS Code, this results in the following display:

This is because

let f _ = ...

is syntactic sugar for:

let f = fun _ -> ...

where fun _ -> introduces a ghost location (the whole fun node).

This is why you needed to disable the check on ghost location in your samples.t test.

Proposed Fix

This PR replaces ghost location handling with the [@merlin.hide] attribute. This allows for more precise control over which parts of the AST should be ignored for inlay hints.

To make this approach work correctly, I refactored the main AST iterator to use Ast_iterators.iter_only_visible. The iterator now ensures attributes are not skipped by recursing on non-immediate sub-terms.

Also, this restructuring simplifies the code and makes inlay hints more consistent across different contexts (e.g., top-level let bindings, class-level let bindings, and expression let bindings).

Additional Changes

• Updated samples.t with new test cases illustrating the improved inlay hints behavior.
• Removed support for the avoid-ghost-location option (both in the CLI and protocol).

Related

• Inlay hints: traverse ghost locations #1892 (an early version of this PR)
• Disable inlay hints under ghost location ocaml-lsp#1290 : the introduction of the ghost location check
• Preliminary inlay hint support ocaml-lsp#1159

Screenshot

[mlasson]

Is it ok to change the protocol/library in a non-backward compatible way ?