`GraphemeCursor::next_boundary()` returns incorrect boundary when chunk starts with ZWJ

https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=8fe7258129406b137529559157edd542

use unicode_segmentation::{GraphemeCursor, GraphemeIncomplete};

#[test]
fn test_graphemes_chunked() {
    use GraphemeIncomplete::*;

    let chunk0 = "👩"; // 4 bytes
    let chunk1 = "\u{200d}🔬"; // 3 bytes + 4 bytes
    let full = &(chunk0.to_string() + chunk1); // 👩‍🔬

    let mut cur = GraphemeCursor::new(0, full.len(), true);
    assert_eq!(cur.next_boundary(chunk0, 0), Err(NextChunk));
    match cur.next_boundary(chunk1, chunk0.len()) {
        Ok(res) => assert_eq!(res, Some(11)),
        Err(PreContext(_)) => {
            cur.provide_context(chunk0, 0);
            assert_eq!(cur.next_boundary(chunk1, chunk0.len()), Ok(Some(11)));
        }
        _ => unreachable!(),
    }
}

the last assert panics and says that it expected Ok(Some(7)), when instead it should be Ok(Some(11))

unicode-rs / unicode-segmentation

`GraphemeCursor::next_boundary()` returns incorrect boundary when chunk starts with ZWJ #118