fix: use tangents for smooth stroke composition

[Kneemund]

• Because of how stroke composition works so far, line segments with highly varying widths had noticeable "dents" in them.
• This was often hidden fairly well by the large amount of path elements, but it becomes much more noticeable with feat: brush stroke simplification #1769.

Before	After

Details

• So far, line segments are constructed as trapezoids with semicircular caps:
  • The trapezoids are constructed to be exactly perpendicular to the stroke segment.
  • They are only continuous with the line caps if both ends have the same width.
• This PR changes the approach, and it can be interpreted in the following way:
  • A circle with the start width is placed at the start of the line segment.
  • That circle is continuously deformed until it reaches the end width at the end of the line segment.
  • This makes the stroke continuous even at different widths. It is achieved by calculating the direction that is tangential to both circles, and connecting the circles along that direction.

Debug View Examples

(The bottom image shows the degenerate case where one circle is contained in the other. It falls back to the previous behavior.)

[Kneemund]

I have a feeling that we could simplify the "direction selection" somehow...

[Doublonmousse]

I wonder if we have a systematic pen up event with 0 pressure giving 0 width at the end here from gtk... Maybe we don't want to have this at the end of each stroke ?

[Kneemund]

Maybe... But this is a more general issue, it's just that rapid pressure changes most likely occur at the end of strokes.