Replace Clipper1 library by Clipper2 library

[rburing]

Reduces release template binary size (by 54.1 kB on Linux).

To-do:

• Resolve TODOs in code
• Use floating point coordinates directly
• Test more extensively

Fixes #90103

[smix8]

You can replace the Clipper XYZ_64 classes with the Clipper D class equivalents. Those do the same except that they allow for direct float input that gets up- and downscaled to the integer grid internally.

So the old Clipper1 manual scaling with SCALE_FACTOR can be removed.

I did a similar upgrade for the 2D navigation mesh baking as an example with PR #89929 here.

The manual Rect clipping in the sprite editor can also be replaced by the "real" RectD Clip of Clipper2 which is far more performant see https://angusj.com/clipper2/Docs/Units/Clipper/Functions/RectClip.htm

[ghmart]

What's about #29886?

[smix8]

What's about #29886?

This PR should imo stay focus on only the upgrade so it can be added without much issues and bike-shed.

Adding new parameters is feature territory as it creates compatibility issues.

It is likely that there is need for a more general rework of the Geometry2D boolean ops later which would be a good place for this and other features. E.g. the single polygon per function limit comes to mind. It is kinda arbitrary as Clipper2 can do way more.

[rburing]

@smix8 Thanks for the tips! I used the D variants of Clipper functions/types now, and used RectClip.

I could not remove the SCALE_FACTOR from Geometry2D since the unit tests expect higher precision.

Should we actually be using Point<real_t> instead of PointD etc.?

Edit: I see I should be using the ClipperD constructor with the precision argument.

[smix8]

Looks good for the Clipper2 replacements.

Not really a 2D and sprite editor user so can't say if there are some subtle differences in the results that could cause some compatibility problems for older projects.

[KoBeWi]

I'm using Geometry2D for merging polygons and this change makes my tool a few times slower. Here's a test scene:
GeometryTest.tscn.txt

This was already discussed in the chat, I thought I'd mention it here for the record.

[rburing]

Here is an image that explains why it was slower:

I have now added

clp.PreserveCollinear(false); // Remove redundant vertices.

to make the resulting polygon the same as it was with Clipper1 (i.e. having just the corner vertices).

Performance is still a bit lower than with Clipper1, but the difference is no longer orders of magnitude.

Edit: In a release build (or a non-dev editor build) there is now no measurable performance difference when running @KoBeWi's project.

[KoBeWi]

In dev build my code runs 10ms slower on average, but it's acceptable in this case.

[akien-mga]

Thanks!