Cycle::winding makes assumptions that don't hold for all cycles

[hannobraun]

Cycle::winding is a method that computes the winding of a cycle (as the name suggests), meaning it determines whether a cycle's half-edges are oriented clockwise or counter-clockwise. The method works well for all cycles made up of straight half-edges (i.e. polygons) and circles. It does not work for all cycles that contain arcs, as it makes some assumptions that don't always hold in the presence of those.

Exhibit A:

fornjot/crates/fj-core/src/objects/kinds/cycle.rs

Lines 46 to 51 in 3e3e992

	let circle = match first.path() {
	SurfacePath::Circle(circle) => circle,
	SurfacePath::Line(_) => unreachable!(
	"Invalid cycle: less than 3 edges, but not all are circles"
	),
	};

Imagine a half-circle whose end-points are connected by a line segment. Totally valid, but unless the arc happens to be the first half-edge in the cycle, it will trigger this panic.

Exhibit B:

fornjot/crates/fj-core/src/objects/kinds/cycle.rs

Lines 61 to 80 in 3e3e992

	// Now that we got the special case out of the way, we can treat the
	// cycle as a polygon:
	// https://stackoverflow.com/a/1165943
	let mut sum = Scalar::ZERO;
	for (a, b) in self.half_edges().pairs() {
	let [a, b] = [a, b].map(|edge| edge.start_position());
	sum += (b.u - a.u) * (b.v + a.v);
	}
	if sum > Scalar::ZERO {
	return Winding::Cw;
	}
	if sum < Scalar::ZERO {
	return Winding::Ccw;
	}
	unreachable!("Encountered invalid cycle: {self:#?}");

The comment is wrong! Imagine a crescent shape. This code treat that like two coincident line segments, meaning sum will be zero and we'll run into the panic.

---

There might be more issues, but those are two that I happened to discover. I believe this once worked as designed, when we were limited to line segments and full circles. Once arcs were introduced, the assumptions here no longer held.

I think it makes sense to hold off on fixing this. Nobody has complained about this so far, and if #2118 works out, fixing it will become trivial.