GDScript: Fix annotation parsing adding new annotation entries

[HolonProduction]

Fixes #98145

The code was indexing the annotation map regardless of the previous check which apparently introduces new entries. So the right error was only emitted for the first parsing but for any subsequent parsing the name would be present in the map.

[dalexeev]

Looks good to me. Please add a test.

[dalexeev]

which apparently introduces new entries

It's probably because of this:

godot/core/templates/hash_map.h

Lines 578 to 593 in 92e51fc

	const TValue &operator[](const TKey &p_key) const {
	uint32_t pos = 0;
	bool exists = _lookup_pos(p_key, pos);
	CRASH_COND(!exists);
	return elements[pos]->data.value;
	}
	TValue &operator[](const TKey &p_key) {
	uint32_t pos = 0;
	bool exists = _lookup_pos(p_key, pos);
	if (!exists) {
	return _insert(p_key, TValue())->data.value;
	} else {
	return elements[pos]->data.value;
	}
	}

For some reason, the non-const overload is chosen, although we meant reading here, not writing. Also, we recently fixed a similar bug for typed dictionaries (see #96797).

godot/core/variant/dictionary.cpp

Lines 86 to 88 in 92e51fc

	// WARNING: This operator does not validate the value type. For scripting/extensions this is
	// done in `variant_setget.cpp`. Consider using `set()` if the data might be invalid.
	Variant &Dictionary::operator[](const Variant &p_key) {

It's a shame that C++ has such a limited and confusing behavior. Maybe we should abandon creating new elements using operator[] in favor of the insert() method? Same for writing and set(), but it will probably be inconvenient.

[HolonProduction]

Done! Really should start making a habit out of adding tests 😅

[Repiteo]

Thanks!