Correctly replace scene root when must_reload in EditorData::check_and_update_scene()

[Rindbee]

We need to update the scene root in multiple singletons to ensure that certain flags are correct. This is what EditorNode::set_edited_scene() does.

Usually we use replace_by to complete the replacement of the scene root. Call EditorNode::set_edited_scene() when the replacing_by signal is emitted to set the new scene root. This is suitable when using a single node to replace, but this may be problematic if the replacing node is a tree. Because during the call to replace_by(), the new node and its child nodes will enter tree during parent->add_child(p_node), and later emits the replacing_by signal. This results in some flags potentially being incorrect during the enter tree of the new node's children.

When the parent scene has to be reloaded because the child scene changes and switches to the parent scene, there is no need to use replace_by() since the scene's diffs are already saved.

Fix #90301.

[akien-mga]

CC @ajreckof

I feel like we're going around in circles there, we modified replace_by for this but now it's no longer needed.
Might be best to go back to the original state and figure out again how to solve the issue without regression.

[AThousandShips]

It does feel like where playing a game of whack-a-bug with this issue, so it'd indeed be good to revert to a clean slate and dig deeper figure out what's causing what, and make sure we have a fix that doesn't cause regression, it seems like some part of the process of replacement here is missed and that it is causing these problems, due to things occuring behind the scenes that were missed

[Rindbee]

It seems inappropriate to use replace_by() inside tree when changing scene root (node type). The replacing_by signal is a bit late.

To prevent certain flags from being wrong when entering the tree due to the cached scene root not being updated. It seems that we need to first let the old scene leave the scene tree, then replace the scene root, and finally use EditorNode::set_edited_scene() to add the modified scene to the scene tree again.

[ajreckof]

I feel really dumb right now 😅 This is in fact what I wanted to do but I don't know why I thought I couldn't directly call it. Obviously some of my previous changes can be reverted but some others are still needed for this to work here is a recap of what should be reverted from my previous commits the rest are needed for this commit.
Rindbee/godot@correctly-replace-scene-root-when-must_reload...ajreckof:godot:reverts-where-needed-keep-where-usefull

[Rindbee]

Maybe the timing I chose for replacing_by in #68599 was inappropriate. Emitting it a little early helps ensure that certain types of nodes are configured correctly when entering the tree as the scene root.

[KoBeWi]

Looks fine.
The third commit should likely be merged with the first one.

[ajreckof]

The third commit won't cause problems in the editor but since it is a signal hence exposed this is still a compatibility breakage. I'm not sure the bug this specifically fix warrant this compatibility breakage. Whatever I think it is better to keep separate if we finally decide to revert it because it was in the end too much of a compatibility breakage.
For us to have a better appreciation of what it fixes and if it is worth the compatibility breakage, it would be great to have a clear reproduction step to a bug that would be fixed by this.
I'm wondering if it shouldn't have it's own PR for those reasons.

[Rindbee]

Create PR #90479 for the third commit.

[akien-mga]

Thanks!