refactor: canvas widget adjustments

[Kneemund]

• Fixes Canvas doesn't stay centered in page after zoom #1720.
• Fixes [Regression] Pinch-to-zoom broken on touchscreen after v0.14.1 #1738.
• Fixes Zoom in: canvas jumps around a bit #1394.
• Applies the workaround from Fix Pinch-to-Zoom With GTK Workaround #1733 to all other zoom gestures and actions.
• Fixes "known issue" described in feat: rotate canvas #1712.
• Fixes gtk_adjustment_configure: assertion 'lower + page_size <= upper' failed log spam by ensuring that the bounds are at least as big as the widget.

---

• Transforms the camera offset into an absolute range for configuring the adjustments of the canvas widget.
• Updates camera offset as soon as adjustments emit a value changed signal, instead of during layout (re-)allocation.
• Autoexpands the canvas when an adjustment changes (like all other gestures/actions), making the autoexpand on edge overshoot redundant. As a preparation for feat: rotate canvas #1712, we cannot rely on edge overshoot callbacks, because when the document is rotated, we have to expand it before reaching the ScrolledWindow edge.
• Because of the above, we hit a GTK bug related to kinetic scrolling. 53af354 works around this bug by temporarily disabling kinetic scrolling when the document is about to be expanded.

---

TODO

• It seems we hit this old GTK bug https://gitlab.gnome.org/GNOME/gtk/-/merge_requests/395. I'm currently trying to implement the workaround used in other apps...

[flxzt]

Looks good, functionally it fixes the outlined issues.
I also like that it fixes this by moving the overshoot logic into the engine,
not relying on GTK's widget overshoot mechanic anymore. This is a cleaner approach.
It also brings down the complexity of this logic a bit (- the needed workarounds) by improving function names, introducing some helpers like surface_to_adjustment and add comments. Which is a big step in the right direction but complexity is still high which I'll need to address in a code cleanup in the future to ensure future maintainability.

[Kneemund]

I agree that this needs to be cleaned up in the future. Especially the surface <-> adjustment conversion should be more hidden/abstracted away, but I couldn't really find a good way to do that within the current code without re-calculating the surface bounds all the time.