fit_to_bezpath_opt panics on a sliver

[mlwilkerson]

Claude found kurbo panics on some curves I'm processing. This is Claude's summary of the issue:

fit_to_bezpath_opt panics on some well-formed, finite inputs. Reproduced on 0.13.0 and 0.13.1.

I hit it through simplify_bezpath with SimplifyOptLevel::Optimize, feeding it flattened polygon-offset contours.

Reproduction

use kurbo::simplify::SimplifyBezPath;
use kurbo::{BezPath, Point};

fn main() {
    // A closed quadrilateral whose four vertices are nearly collinear.
    let pts = [(1.0, 0.7), (4.1, 2.0), (4.4, 2.1), (0.0, 0.0)];

    let mut path = BezPath::new();
    path.move_to(Point::new(pts[0].0, pts[0].1));
    for (x, y) in &pts[1..] {
        path.line_to(Point::new(*x, *y));
    }
    path.close_path();

    let _ = kurbo::fit_to_bezpath_opt(&SimplifyBezPath::new(path), 2.0); // panics
}

thread 'main' panicked at kurbo-0.13.1/src/fit.rs:646:61:
called `Option::unwrap()` on a `None` value

(fit.rs:650:61 on 0.13.0.) Expected: a BezPath, or a graceful fallback — not a panic.

Cause

The panic is the last statement of fit_to_bezpath_opt_inner:

let (c, _) = fit_to_cubic(source, t0..t1, accuracy).unwrap();

fit_to_cubic returns Option and is treated as fallible at every other call site in fit.rs. Here it's unwrapped,
assuming a segment that fit_opt_segment accepted at the smaller per-segment budget x must also fit at the larger
accuracy.

It need not, because fit_to_cubic changes strategy rather than just relaxing as accuracy grows:

if chord2 <= acc2 {
    // Special case very short chords; try to fit a line.
    return try_fit_line(source, accuracy, range, start.p, end.p);
}

A bigger accuracy makes this branch more likely, and try_fit_line returns None when the curve strays further
than accuracy from the chord. So the trigger is a sub-range with arc length ≫ accuracy but chord ≪ accuracy —
geometry that doubles back on itself, which is what a nearly-degenerate polyline is full of.

Notes

• It's an accuracy window, not a threshold: this input panics at accuracy = 2.0 but succeeds at 0.5, 1.0, 4.0 and
  8.0.
• fit_to_bezpath handles the same input at every accuracy above, so it's specific to the _opt path.
• The input is minimal — no three-vertex subset reproduces, the unclosed form doesn't, and only this starting vertex
  does.

Possible fix

Stop unwrapping and give the loop somewhere to go when a sub-range can't be fit — fall back to the chord, or propagate
the failure — matching how the other fit_to_cubic call sites already treat it.

Repro

kurbo_fit_opt_panic.rs.txt