AsObjectArg trait enabling implicit conversions for object parameters

[Bromeon]

Adds a trait AsObjectArg<T>.
Godot engine functions that have a parameter impl AsObjectArg<T> will now accept:

• Gd<U> where U: Inherits<T>
• &Gd<U>
• Option<Gd<U>>
• Option<&Gd<U>>
• Gd::null_arg(), a special expression only valid for null arguments

Closes #156.
Closes #796.

Needs to land in v0.2 because some call-site patterns like node.clone().upcast() are now ambiguous and no longer possible.

---

Older message

To be decided:

1. Should we impl AsArg<T> for Gd<T> (i.e. values, not just references)?
   • allows destructive move
   • creates possible move errors that need a lot of fine-tuning, instead of just always passing &obj
   • can create more monomorphizations (especially if also both Option<Gd<T>> and Option<&Gd<T>> are supported)
2. What is a good inner type?
   • RawGd creates lots of copies, and may be unsafe/unsound to cast, plus needs destructor deactivated
   • raw object pointers that can be null?
   • dedicated ArgView type wrapping a pointer?

[Ughuuu]

Can't comment much as I don't really understand what this will achieve. With this I see you are not calling explicitly .clone anymore in some cases. Does it still clone in those cases or not anymore?

[Bromeon]

Does it still clone in those cases or not anymore?

Possibly in the initial implementation, but my plan is definitely to get rid of any unnecessary clones. Since this would then be an internal perf improvement, it might be a separate PR.

[lilizoey]

I think we could just use Option<&__GdextObject>? (__GdextObject is the opaque type we generate for use in GDExtensionObjectPtr etc.) This would effectively be a nullable pointer with a lifetime. So we can more easily avoid use-after-free and similar. While keeping the cheapness of passing around a pointer. Ideally this would use extern types but that's not been implemented for a while now.

This would probably force the trait to be an unsafe trait though, since it'd effectively erase the type. But it should also simplify implementation a fair bit.

[Bromeon]

The problem with Option<&__GdextObject> (or any new type wrapping a pointer) is that arguments are passed in to the signature tuple, so we'd need to implement again all the traits that RawGd has: ToGodot/GodotConvert/GodotFfi/GodotNullableFfi/GodotFfiVariant/GodotType/...

I might see how bad it is, maybe we can reuse code somehow.

[GodotRust]

API docs are being generated and will be shortly available at: https://godot-rust.github.io/docs/gdext/pr-800

[Bromeon]

Also declared bugfix, since it fixes #156 which wasn't previously possible.