GDScript: Fix subscript resolution on constant non-metatype GDScript base

[dalexeev]

• Fixes Regression in "GDScript: fix segfault on invalid script" with const foo : GDScript= preload(...) #92534.
• Resolves (partially) Parser error occurs when attempting to access constant or static members of statically typed preloaded constants. #89077.

#92035 exposed a bug with the static analyzer tries to get GDScript properties when they might be invalid (now there is a guard against it). This is due to the fact that when reducing subscript on a constant base, we use Variant::get_named(), while we should use the analyzer. In #79510, I fixed a related bug, but I didn't take into account that a constant value could have a broader type due to type hints and casting.

We also have a quirk that GDScript classes are both types and values. What causes collisions (#75392, #76414) is whether you mean a static class member or a non-static member of the GDScript resource (for instance, resource_path). Also, we don't have first class types and the is_meta_type flag sometimes leads to strange bugs.

Some users add unnecessary type hints to const-preload, which removes metatypes.

const MyClass: GDScript = preload("./my_class.gd")
var test := MyClass.InnerClass.new()
#                   ^^^^^^^^^^
# The specified type `GDScript` has no member `InnerClass`,
# but the value has and this is known in advance.

It might make sense to distinguish between a GDScript class as a type and as a value, but we proceed from the current state of affairs and strive to maintain backward compatibility. Now, for a constant GDScript base without the is_meta_type flag, the static members of the class are checked first, and then the non-static members of the GDScript class. For GDScript with the is_meta_type flag, only static class members are still checked.

[akien-mga]

Makes sense to me. Thanks for the quick fix, and the detailed explanations!

[akien-mga]

Thanks!