Remove error condition from get_global_transform()

[KoBeWi]

Removes the arbitrary error in get_global_transform() method.

The error makes no sense for a couple of reasons:

• it only triggers in debug build, which means the method works differently in release build
• even with the error, the transform may not be truly "global", because Node ancestors stop transform propagation
  • if node is not inside tree, it has null parent, which is effectively the same as having Node parent

Closes #30445

[Sauermann]

This change makes sense to me. Don't see any problems with the implementation.

[aaronfranke]

Makes sense to me.

[mhilbrunner]

@reduz originally added this in 83ae9a5, thoughts?

[Sauermann]

@mhilbrunner
In 83ae9a5 CanvasItem::set_notify_transform is changed and adds a call to get_global_transform, but verifies beforehand, that the node is in the tree.
So this PR does not change anything for CanvasItem::set_notify_transform, as far as I can tell.

[akien-mga]

Thanks!

[timothyqiu]

Cherry-picked for 3.6.

[lawnjelly]

Just seen this has been cherry picked! Yay, had been waiting for this for a while (see #30445 ) .

Ah, I've just seen, this PR is for canvas item, and the issue is for Spatial.

Also a possible bug - I haven't checked whether this occurs yet. If you remove a branch from the scene tree then unless the global transform is made dirty then the results will be stale (i.e. still represent the transform within the tree). So when removing from the tree we would need to propagate dirty flags to ensure consistent result. I'm not sure if this is done already, will check.

EDIT: The global_invalid is set when exiting the tree, so it should be consistent and there is no bug there currently. 👍
This should be checked though if we decide to do same for 3D.