Avoid crash on exiting due to late prints

[RandomShaper]

Fixes #80152.

UPDATE: Probably fixes #79379.
Bugsquad edit: And fixes #79506.

The MRP there adds a message when the EditorNode has already been destroyed. The most obvious fix is done here, which is just validating the singleton when a deferred call happens. The problem is that static calls in the message queue can't be validated.

Now, there are potential alternatives to this fix that may be done instead of in addition:

a) Let the static callables have some sort of custom validation.
For instance, there's this call:
callable_mp_static(&EditorNode::_print_handler_impl).bind(p_string, p_error, p_rich).call_deferred();
With this approach it could become something like this:
callable_mp_static(&EditorNode::_print_handler_impl).bind(p_string, p_error, p_rich).add_validator([]() -> bool{ return EditorNode::get_singleton(); }).call_deferred();

b) Like a), but using a sentinel pointer:
callable_mp_static(&EditorNode::_print_handler_impl).bind(p_string, p_error, p_rich).add_validity_sentinel(&EditorNode::singleton).call_deferred();

c) Make all deferred calls of this kind via the object instance instead of static. That would enable the current validated call mechanism to do its magic.

[AThousandShips]

Does this fix #79379?

[RandomShaper]

Does this fix #79379?

Very likely.

[akien-mga]

Makes a lot of sense to cleanup the singleton like this.

[akien-mga]

CC @KoBeWi for the discussion on static callable validation - but this PR is ready to merge as is IMO.

[KoBeWi]

If a static function depends on an instance, it shouldn't be static, no?
Also if it's expected that the instance it involves can be invalidated, the check should be inside the function itself I think.

[akien-mga]

Still seems good after changes.

[YuriSizov]

Thanks!