Fix updating cached singletons when reloading GDScripts

[DmitriySalnikov]

Fixes #85004

Since GDScriptLanguage::reload_all_scripts() is always called when the GDExtension reload is completed, I decided to update the singleton cache there.
I used the original singleton caching code, added a key existence check, and just overwrite the values.

Without globals.has(E.name) singletons that were registered via GDScript itself or for example GDScriptLanguageProtocol can be added to the cache after the GDExtension reload and used as the rest of the singletons in GDScript.

#52162 could be fixed if calling reload_all_scripts somewhere after loading the editor (there?) and remove globals.has(E.name). I tried updating scripts in Engine::add_singleton, but it is advisable to do it deferred, and Engine is a regular class that cannot call methods via callable_mp (although it was possible to add a static method and a static boolean variable to prevent multiple calls in one frame). This can be tested in a separate PR, but I'm not really interested in it.

It would be cool to merge this before the 4.2 release.

[DmitriySalnikov]

I simplified this commit. Now everything happens in reload_all_scripts. Singletons are still updated only in the editor. But I'm not sure if it's necessary to place this code inside a block with a lock?

Right now there are only two cases when a reload_all_scripts is called: RemoteDebugger in a running game, after saving scripts in the editor, GDExtensionManager in the editor when the libraries are reloaded. Therefore, only the second case will be affected here, while nothing is added or deleted, only pointers are updated, which without this change do not allow using singletons with GDExtension hot reload.

Btw 4.2 will be the first release with hot reload support for GDExtension. 4.2 RC2 isn't the last release candidate, is it?

[akien-mga]

Btw 4.2 will be the first release with hot reload support for GDExtension. 4.2 RC2 isn't the last release candidate, is it?

A release candidate should always expect to be the last, otherwise it's not a release candidate ;)

This doesn't seem critical (doesn't prevent shipping games, only impacts hot reload during extension development), so it doesn't seem urgent to merge mere days before the stable release. Like other bug fixes, it can be merged afterwards and cherry-picked for 4.2.1.

[ValorZard]

I'd like to offer a counterpoint. Hot Reloading for GDExtension seems like it will be a very popular feature for 4.2, which means making sure its bug free and ready to be used is probably important. Although, I'm not sure how annoying this bug is, so YMMV.

[dsnopek]

Thanks!

This fixes the issue with the MRP from #85004 in my testing.

The code makes sense to me based on looking at the references from the PR description, but it should probably be reviewed by someone on the GDScript team.

[vnen]

Makes sense to me. Refreshing the pointers to the singletons are necessary if the extensions are reloaded.

[akien-mga]

Thanks!

[YuriSizov]

Cherry-picked for 4.2.1.