Improve editor state persistence

[Geometror]

This PR is an attempt to improve the editor state preservation when restarting the editor. Strangely I couldn't find any GH issues on this topic, but I know this has been a inconvenience for quite a few users, myself included :)

The following settings are now correctly saved and loaded:

• Active open scene
• Bottom panel:
  • Height (split offset) offset
  • Active tab (e.g. debugger, shader editor)
  • Shader editor:
    • Open shaders (incl. VisualShader)
    • Selected(currently editing) shader
  • Active debugger tab
• Active tab per dock
• Active main editor plugin (2D/3D/Script/Asset/...)
• Script editor:
  • Selected(currently editing) script
• Selected nodes per scene
• Some viewport specific settings which were not correctly preserved (mostly regressions)
  • Orthogonal/perspective view
  • Advanced display (debug) modes (e.g. normal buffer, GI buffer, shadow map)
  • Half resolution
• FileSystem dock
  • selected paths
  • expanded folders

Additional misc changes (internal, not compat-breaking):

• Rename key "use_orthogonal" to "orthogonal"
• Rename VIEW_DISPLAY_SHADELESS to VIEW_DISPLAY_UNSHADED
• Some renames of internal methods and variables to better describe what they do

While this PR should fix the most annoying cases where the layout/settings would be lost, there are almost certainly still some left. Please let me know if you find one that should be included in this PR :)

Somewhat related side note: I realized that after 4.0 is out we need to do some significant internal refactoring of some core editor classes/files (e.g. editor_node/node_3d_editor_plugin). There's lots of duplicated/legacy code, confusing/short variable names, etc. which make it very hard to maintain them.

And lastly, since some visual comparisons won't do any harm:

---

Editor before and after the restart with this PR

Editor after restart with beta 16 (manually switched from scene "2d_test" to "main_test")

Bugsquad edit: Closes godotengine/godot-proposals#4789

[Paulb23]

Strangely I couldn't find any GH issues on this topic

The ShaderEditor is tracked by #29094. Though, that would include caching the full sate i.e bookmarks, carets etc that the ScriptEditor currently does.

[Geometror]

Updated the PR with the following changes.

• For the current scene, selected script and selected shader the resource path is now stored instead of the tab index. (as suggested by @Paulb23)
• Add editor setting "editors/shader_editor/behavior/files/restore_shaders_on_load" (default: true)
• Add editor setting "text_editor/behavior/files/select_scene_root_script_on_load" (default: false) [This might require some discussion. Do you find this behavior useful?]

[michaelhue]

It would be great if the FileSystem panel could preserve expanded folders.

[Geometror]

Rebased and updated:

• Fixed a crash
• Improved state preservation for the FileSystemDock
  • selected paths
  • expanded folders (as suggested by @michaelhue)

[Geometror]

• Fixed a bug that caused the selected nodes not to be restored when the scene's root node was a canvas item.
• Some internal renames of unclearly named methods

[KoBeWi]

There's one more bug with filesystem dock. It stores <null> in uncollapsed paths (probably from res://).

EDIT:
It's not res://, because when I collapse it the null is still there 🤔
Another thing is that if you have uncollapsed directory, which is a child of a collapsed one, it's not stored.
Also, probably not that important, but if you select a file and then collapse the parent directory, it will be uncollapsed on launch, because it navigates to the selected file.

[Geometror]

Okay, rewrote a few things.
@KoBeWi All problems you mentioned regarding the FileSystemDock should be gone now.
I also fixed a bug where an uncollapsed directory, which is a child of a collapsed directory, got collapsed when you saved the scene (and the Tree is rebuilt).

FileSystemDock::_compute_uncollapsed_paths() is renamed to FileSystemDock::get_uncollapsed_paths() const , made public (so it can be used in EditorNode) and uses the new BFS algorithm (which does not include non-directory paths and fixes the bug described above).

In addition, I introduced get_prev_in_tree(bool p_wrap) and get_next_in_tree(bool p_wrap) (together with some minor internal refactoring).
These were necessary because the TreeItem API lacked a method like get_next_visible for iterating over the tree via a DFS, which includes all items, not just the visible. get_next just returns a TreeItem's sibling.

(As an alternative approach, we could add two parameters (default: false) to get_next, one for changing the mode to DFS and another for wrapping)

[KoBeWi]

Needs rebase 🙃

[Geometror]

Floating docks should now be restored correctly again, somehow a small code section got automatically removed during the rebase. Addressed all other review comments too.

[KoBeWi]

There's one last documentation comment left, but other than that I think it's fine now.

[akien-mga]

Thanks!