Ensure that SceneTree is initialized and finalized at correct time

[RedwanFox]

SceneTree should be fully initialized before any tree operation with any node and finalized only after all nodes exited tree.

Accounts only for in-tree operations, not constructors. After fix init-deinit order will be:

1. constructors (_init) in unspecified order
2. SceneTree native _initialize
3. SceneTree script _initialize
4. All nodes in-tree lifecycle (_enter_tree ->_ready -> ...-> _exit_tree)
5. SceneTree script _finalize
6. SceneTree native _finalize

Partially fixes #71695

[akien-mga]

Possibly related to #70771 ? CC @kleonc

[RedwanFox]

Added one more pull request as follow up to this #72266
It's experimental and is meant for discussion.

[kleonc]

I agree with the changes, it's logical to:

• in SceneTree::initialize firstly initialize the base class (call MainLoop::initialize), then do SceneTree-specific initialization (set tree for the root),
• in SceneTree::finalize firstly finalize the SceneTree-related stuff, then finalize the base class (call MainLoop::finalize).

It's the same ordering as for constructors/destructors.

[kleonc]

Possibly related to #70771 ? CC @kleonc

@akien-mga Kinda related as both are fixing some aspects of likely not-so-logical initialization order for MainLoop/SceneTree. But I do think both PRs are fine on their own / independent.

[RandomShaper]

Looks good for early testing in 4.2.

[YuriSizov]

Needs a rebase though :)

[RedwanFox]

@RandomShaper Could you also look into this #70771 ? It would be nice to have both fixes in upstream
@YuriSizov done

[kleonc]

As already mentioned in #72248 (review) this initialization/finalization order makes more sense to me.

Can't think of anything this change could break besides the cases where user implements custom MainLoop extending SceneTree, and somehow relies on the previous ordering of these callbacks. I find this unlikely though.

Looks good for early testing in 4.2.

☝️

[YuriSizov]

Thanks! Yeah, we're aiming to merge this for 4.2 dev1.

[YuriSizov]

Thanks!