Instance creation now goes through proper placeholder process

[Bromeon]

Fixes #1404.
Fixes #912.

This PR changes Gd::default_instance() so it goes through the editor/placeholder process when needed.

I decided against blanket call to ClassDb::instantiate(), because this downgrades an indirect C function call (create callback) to a Godot reflection call with all the bells and whistles + extra Variant marshalling. I didn't want to have this overhead on hot paths in games, as object default-creation is still a pretty fundamental operation.

Instead, the code first checks whether editor is running, and only then goes through ClassDb::instantiate(). For non-editor runs, the old way works fine, so we keep using that.

Checking whether editor is running turns out to be harder than I thought, and takes up half of this PR (but contained in last commit). The problem is that Engine class isn't available in init-stage < Scene for Godot < 4.4, and I've found no other easy way than Engine::is_editor_hint() to check this. So we need to have a tristate (editor, runtime, unknown).

---

Apart from this, the PR adds quite a few itests around placeholder behavior. Thanks to #1591, we can now test those and will be notified immediately if we break something or Godot changes semantics.

I would use this PR as a base for future improvements, e.g. no longer implying tool in #[class(singleton)].

[GodotRust]

API docs are being generated and will be shortly available at: https://godot-rust.github.io/docs/gdext/pr-1597

[TitanNano]

Looks good. I also tested it with my project, and it didn't break any of my tool classes. 👍 Maybe we provide a deprecated alias for is_editor as is_editor_hint? That is currently a breaking change.

[Bromeon]

Oh, very good catch, I missed that we publicly re-export this. I'll add it back, we can always change in the future.

[Yarwin]

why not https://godot-rust.github.io/docs/gdext/pr-1597/godot/init/fn.is_editor.html ?

[Bromeon]

#1597 (comment) 🙂 but yes.

[Yarwin]

We don't use "runtime class" anywhere in public-facing docs. Can we use tool class instead, like earlier? In the past you raised argument that it is implementation detail which might not be obvious/intuitive to user (since Godot doesn't really use it anywhere)

[Bromeon]

Fair, I changed it to "non-tool class".

Maybe we can at some point introduce "runtime class" as a concept, but it should probably be in a more prominent place, not some isolated error messages.

[Yarwin]

I checked with simple setup in editor - init creates real instances (as it should), editor operates on placeholders:

Details

#[derive(GodotClass)]
#[class(init, base = Resource)]
struct MyClass {
    #[export]
    #[init(val = GString::from("my default"))]
    #[var(get=my_something)]
    something: GString,
}

#[godot_api]
impl MyClass {
    #[func]
    fn my_something(&self) -> GString {
        godot_print!("henlo");
        self.something.capitalize()
    }
}

#[derive(GodotClass)]
#[class(init,  base = Node)]
struct MyClass2 {
    #[export]
    #[var(get = my_get2)]
    value: i32,
    #[export]
    other_value: Option<Gd<MyClass>>,
    base: Base<Node>,
}

#[godot_api]
impl MyClass2 {
    #[func]
    fn my_get2(&self) -> i32 {
        godot_print!("hello {:}", self.to_gd());
        42
    }
}