Fix unrestricted mouse-event propagation to SubViewports for Physics-Picking

[Sauermann]

resolve #17326
resolve #58902
resolve #75015
resolve #75019

previous description

Currently, Events get processed by Nodes in order of the green arrow in the following graphic, when aSubViewportContainer is within the Root Viewport:

The red "X" denotes the place where set_input_as_handled() is called within Viewport::_gui_input_event, when an InputEventMouseMotion happens over the SubViewportContainer. Here is the according location in the source code:

godot/scene/main/viewport.cpp

Lines 1789 to 1796 in 84f2c68

	bool stopped = false;
	if (over && over->can_process()) {
	stopped = _gui_call_input(over, mm);
	}
	if (stopped) {
	set_input_as_handled();
	}

If the SubViewportContainer marks the event as stopped, _unhandled_input is never called.
If the SubViewportContainer ignores the event, it is processed in _unhandled_input, even though ignoring should be a synonym for not sending the event to the SubViewport of SubViewportContainer.

This pull request makes sure, that the event reaches _unhandled_input inside the SubViewport before the "X":

This is done by moving the push_unhandled_input call from Window::_window_input to Viewport::push_input.

This change makes SubViewportContainer::unhandled_input redundant, since it would distribute the event to _unhandled_input inside the SubViewport a second time.

Shortcut Events now need to be distributed via push_input, in order for them to be able to reach SubViewports.

MRP for all linked issues: SubViewportEventBugs.zip

Currently InputEvents get sent unconditionally to SubViewports via SubViewportContainer::unhandled_input, even if the event-position is outside of the area of the SubViewportContainer. This leads to problems like #58902, where the mouse enters CollisionObjects2D, even if they are not visible. (in the MRP, clicking in the red area causes a mouse-click on the non-visible Logo within the SubViewport).

Also hovering CollisionObjects2D within SubViewports currently is reversed:

• If the SubViewportContainer is set to MOUSE_FILTER_IGNORE, the CollisionObject2D does receive mouse events.
• If the SubViewportContainer is set to MOUSE_FILTER_STOP, the CollisionObject2D doesn't receive mouse events.

A third bug is that a Panel with MOUSE_FILTER_STOP, parent of a SubViewportContainer, prohibits Physics-picking in the SubViewport.

The fourth bug, that gets fixed is that TouchScreenButtons don't receive some mouse events within _unhandled_input. (TouchScreenButton doesnt send event via _unhandled_input)

This PR solves all problems by moving the call of push_unhandled_input from Window::_window_input to Viewport::push_input.
This makes sure, that only events with a position within the area of the SubViewportContainer are considered for physics picking.
This change makes SubViewportContainer::unhandled_input redundant, since it would distribute the event to _unhandled_input inside the SubViewport a second time.

Updated 2022-09-26: fix merge conflict, made sure that this PR is still relevant after #61088 and adjusted description.
Updated 2022-10-14: adjust shortcut event handling
Updated 2022-11-06: update documentation of push_input.
Updated 2022-11-23: fix merge conflict
Updated 2023-01-27: fix merge conflict, update description based on the recently merged #58334
Updated 2023-03-17: fix merge conflict, updated MRP, linked a third bug, that this PR fixes.
Updated 2023-05-08: linked a fourth bug, that this PR fixes and updated description and MRP.

[KoBeWi]

Tested the attached project + some random stuff and it appears to be working correctly.

[RandomShaper]

I think it's a good time to merge this.

[akien-mga]

Thanks!

[WolfgangSenff]

I commented already in #17326, but I think some or much of this PR may be missing from the latest beta, 4.2.beta5. At least, I looked at the code and it seemed like some of it was missing - could be it was updated in the future and I missed that.

[Sauermann]

@WolfgangSenff The only thing, that changed, was the renaming of _push_unhandled_input to _push_unhandled_input_internal, but the important parts of this PR are still in the code.