[3.x] Include the follow-viewport-transform into CanvasLayer transform calculations

[Sauermann]

The follow-viewport-transform was missing from several calculations

fix #35965 for 3.x
3.x version of #59682

A dedicated backport was necessary, because of merge conflicts

[akien-mga]

Thanks!

[kleonc]

@akien-mga I guess there's nothing against backporting this to 3.5? 🤔 Follow viewport is still buggy in there.

[akien-mga]

Sure. Might have to wait for 3.5.3 unless you're very confident in it, as I intend to release 3.5.2-stable soon now that it's had 2 RCs without regressions reported.

[kleonc]

Sure. Might have to wait for 3.5.3 unless you're very confident in it, as I intend to release 3.5.2-stable soon now that it's had 2 RCs without regressions reported.

I'm confident about the correctness of this PR. If anything, I think backporting this could break someone's code only in case they'd be manually incorporating the relevant transform as a workaround for the previous buggy behavior. But if breaking the workarounds by fixing the original bug is fine, then I see no reason to postpone it to 3.5.3, can be done in 3.5.2 as well.

For reference just saw someone still having a problem with this: https://www.reddit.com/r/godot/comments/10ly7dz/get_global_mouse_position_not_accurate_when/.

[akien-mga]

Cherry-picked for 3.5.3.