GDExtension: create_instance_func can be NULL to inhibit construction from GDScript

[Bromeon]

This is a small quality-of-life change that's useful for classes which should either not be instantiated (e.g. static methods) or only instantiated from the GDExtension binding. The GDScript constructor call MyClass.new() now causes an error, which has the nice side effect that forgetting to specify create_instance_func leads to a clear error rather than a half-constructed object.

Currently, the code already checks for the condition:

(ti->native_extension && !ti->native_extension->create_instance)

so the API header comment /* this one is mandatory */ is wrong. However, the ClassDB code doesn't act upon the null pointer, instead it constructs a "zombie" instance with initialized base but uninitialized extension.

---

The check at the bottom of the function,

if (ti->native_extension && ti->native_extension->create_instance)

could now theoretically be simplified to:

if (ti->native_extension)

however I'm not sure if that's desired, as another refactoring might quickly introduce UB (null pointer dereferencing).

[Bromeon]

The 2nd commit reuses duplicate code in a common method.

[Bromeon]

Fixed a few issues, encountered a null pointer dereference of obj in the test runner in this CI run:

godot/modules/gdscript/tests/gdscript_test_runner.cpp

Lines 558 to 565 in 2116318

	// Create object instance for test.
	Object *obj = ClassDB::instantiate(script->get_native()->get_name());
	Ref<RefCounted> obj_ref;
	if (obj->is_ref_counted()) {
	obj_ref = Ref<RefCounted>(Object::cast_to<RefCounted>(obj));
	}
	obj->set_script(script);
	GDScriptInstance *instance = static_cast<GDScriptInstance *>(obj->get_script_instance());

I'd need to check why this happens exactly. The first commit passed CI, so probably I introduced a mistake during the refactoring.

Edit: fixed, bad parentheses in boolean condition

[Bromeon]

There's a pending related discussion:

Can we support a void *user_data argument in create_instance_func, providing different ways to construct an instance from GDExtension? If yes, the create_instance_func could also return a null pointer to signal "not constructible this way" (e.g. no default constructor).

[Bromeon]

Oof, sorry for the excessive tagging -- I merged latest master and it included all commits from people in the meantime 😕

Should be fixed now, but unfortunately the review requests don't go away.

[groud]

PR review meeting: may be superseeded by #66979. If that's the case, we could close this PR.