Draw borders with the same color in single bezpath.

[AustinMReppert]

This is a small fix to ensure adjacent borders with the same color are drawn in a single bezpath to avoid artifacts.

Gemini was used to assist with the code.

Fixes #359

prior:

Of tests run:
201 tests CRASHED (0.77% of run; 0.31% of found)
12020 tests PASSED (45.93% of run; 18.48% of found)
13950 tests FAILED (53.30% of run; 21.45% of found)

with border change:

Of tests run:
201 tests CRASHED (0.77% of run; 0.31% of found)
12130 tests PASSED (46.35% of run; 18.65% of found)
13840 tests FAILED (52.88% of run; 21.28% of found)

Prior:

With border Fix:

[nicoburns]

This can be simplified:

let mut next_border_index = start_border_index + 1;
let mut border_path = borders[start_border_index].1.take().unwrap();
while next_border_index < count && borders[next_border_index].0 == color {
    border_path.extend(&borders[next_border_index].1.take().unwrap());
    next_border_index += 1;
}
scene.fill(Fill::NonZero, self.transform, color, None, &border_path);

The key insight here being that the has_multiple_edges "if" is using the same condition as the while loop in the first branch of the if. So we simplify to only using a while loop that just does zero iterations in the "else" case.

[nicoburns]

The docs on BezPath::extend (https://docs.rs/kurbo/latest/kurbo/struct.BezPath.html#method.extend) say:

Note that if you’re attempting to make a continuous path, you will generally want to ensure that the iterator does not contain any MoveTo or ClosePath elements. Note especially that many (open) shapes will start with a MoveTo if you use their path_elements function. Some shapes have alternatives for this use case, such as Arc::append_iter.

Which I think we are doing here?

I guess if it works this is fine. But it would be nice to understand if/why this is ok.

[nicoburns]

Nit: It might be nice for readability to use a struct for this? So we have .color and .path rather than .0 and .1...

[nicoburns]

Gemini was used to assist with the code.

I appreciate the disclosure!

[behdad]

Does this have any performance implications? In hb-raster, Claude prototyped a fix for this issue, but the performance degradation was too high (~50%).