[3.x] Don't terminate library if other GDNatives are still using it

[Waridley]

This PR is being made directly to 3.x because there are apparently so many massive changes to GDNative coming in 4.0 that I have no way to test this or even know if it will be relevant to 4.0.

If a GDNative library is marked as reloadable, GDNative::terminate() checks GDNativeLibrary::loaded_libraries to see if any other GDNative objects are currently using the library before actually closing it, but they weren't actually being added to the list beyond the first one.

Now that NativeScriptLanguage::unregister_script() calls GDNative::terminate() when NativeScripts are left referencing the library (introduced by hot-reloading in 311ca0c), this means that libraries can be terminated out from under GDNative objects just by freeing an object that happens to hold the last NativeScript using a library.

This could happen previously by creating two GDNative objects referencing the same library, and then terminating one of them, but now it's relevant to the more widely-used NativeScripts as well. This is still a relatively rare scenario, using the same library for both a basic GDNative object and creating NativeScripts from it, but this is exactly what the godot-rust tests do to verify that the bindings work both with and without NativeScript functionality.

There is one issue with this fix, which makes it technically a breaking change: without this fix, dropping the last NativeScript terminates the library, and creating a new one would re-initialize it; while with this fix, godot_nativescript_init can end up being called multiple times without terminating the library. Since there is no godot_nativescript_terminate counterpart, libraries have no way of knowing that they can do any NativeScript-related cleanup until godot_gdnative_terminate is called. However, if they were expecting godot_gdnative_terminate to be called before godot_nativescript_init was called again, invariants could be broken and lead to crashes or even UB on the next godot_nativescript_init callback.

The workaround, whether this fix is applied or not, is to mark the library as not-reloadable (luckily this case was fixed by 5228871). However, since the current situation can very easily lead to undefined behavior, a solution needs to be decided on, and the changes in this PR appear to be the original intended behavior.

[Waridley]

I think this should probably be merged before 3.x is abandoned. The only ways to mitigate the issue are to forfeit reload ability or maybe to keep a dummy NativeScript around to prevent unloading. There are other ways to detect and mitigate the re-calling of godot_nativescript_init

Maybe this will help make it clearer what's going wrong:

now:

• Library initialized
• GDNative object created pointing to library
• NativeScript instance created
  • Allocates another GDNative object pointing to the same library
  • godot_nativescript_init called
• NativeScript instance dropped
  • 1 GDNative object freed, but 1 remains
  • No more NativeScripts
    • Library unloaded
• Another NativeScript created
  • Undefined Behavior (SEGFAULT at best)

With this fix:

• Library initialized
• GDNative object created pointing to library
• NativeScript instance created
  • Allocates another GDNative object pointing to the same library
  • godot_nativescript_init called
• NativeScript instance dropped
  • 1 GDNative object freed, but 1 remains
  • No more NativeScripts
    • Library NOT unloaded due to remaining GDNative object
• Another NativeScript created
  • Allocates another GDNative object pointing to the same library
  • godot_nativescript_init called again
    • Potentially UB, but preventable by library

[akien-mga]

CC @godotengine/gdextension @Bromeon

[AThousandShips]

Suggested change

	// in case user triggers a call to terminate in the callback below
	// In case user triggers a call to terminate in the callback below.

Per the comment style.