Fix node processing order

[RandomShaper]

Node processing used to work like this (disclaimer: pseudocode):

FOR-EACH (Physics, Idle):
	FOR-EACH node:
		node.internal_process()
	FOR-EACH node:
		node.process()

After #75901 it changed to (note changed nesting of internal/external and node iterations, as well as the relative order of internal/external):

FOR-EACH (Physics, Idle):
	FOR-EACH group:
		group.message_queue.flush()
		FOR-EACH node in group:
			node.process()
			node.internal_process()
		group.message_queue.flush()

This PR changes the processing so it looks much closer to what it was like. Specifically, while still supporting group processing, it makes internal happen first and changes the nesting of internal/external and node iterations:

FOR-EACH (Physics, Idle):
	FOR-EACH group:
		group.message_queue.flush()
		FOR-EACH node in group:
			node.internal_process()
	FOR-EACH group:
		FOR-EACH node in group:
			node.process()
		group.message_queue.flush()

UPDATE: See #78745 (comment).

Bonus: Fix issues in group-processing related setters.

WARNING: While this fixes a relevant compat breakage, it's very likely that we have edge cases here and there. Moreover, this would need to be smoke and non-smoke tested to be sure the remedy is not worse than the illness.

Fixes #77548.

[akien-mga]

Tested and confirmed this fixes #77548.
I'll try to find some other projects to make sure there's no obvious breakage.

[reduz]

While this is a solution, I am pretty sure that this will have a very significant performance impact, specially when having to deal with thread groups. So I am not very sure that doing this change just to warrant a specific processing order is better vs simply assuming that processing order will not be there any more.

Does not simply changing it to this:

FOR-EACH (Physics, Idle):
	FOR-EACH group:
		group.message_queue.flush()
		FOR-EACH node in group:
		        # just swap these
			node.internal_process()
			node.process()
		group.message_queue.flush()

Resolve the problem? Otherwise, how could this be solved in the context of the skeleton so we don't have to do such major change with the performance implications it carries with it?

[reduz]

To further clarify, with this PR, threads will most of the time be organized, allocated, started and waited to do essentially nothing, so I think this is not the right solution to the problem.

This could be optimized by improving the thread allocation in the process code to avoid them starting if they are going to do nothing, but I think this is a ton of work with little benefit, all for keeping an order of processing that is not very important in the remaining 99.99% of nodes or for Godot users.

So I insist that I would find a solution to the problem elsewhere without doing this type of change.

[RandomShaper]

It seems that switching the order of internal/non-internal does the trick for the bug, so I've updated the PR removing the other changes.

This is the original patch (as a backup, just in case): d.txt

[akien-mga]

That's a more minimal and safe change in line with what reduz asked for, so I'm relatively confident approving this, TIWAGOS of course.

[akien-mga]

Thanks!