Drop the unicode-width bound once ratatui counts halfwidth sound marks once #718

Goal

Drop the unicode-width bound once ratatui counts each halfwidth katakana sound mark (U+FF9E, U+FF9F) as one cell with unicode-width 0.2.3 or later, so ox picks up new width tables again and measures these marks the way ratatui paints them.

ratatui-core 0.1.2 adds a cell to each mark in CellWidth (ratatui/ratatui#2499) because unicode-width used to count them as zero-width. unicode-width 0.2.3 counts them as one cell itself (unicode-rs/unicode-width#89), so with both, ratatui lays ｶﾞ out over three cells while terminals draw it in two. #711 hit this in testing::tests::buffer_rows::read_an_offset_area_and_skip_the_cells_wide_symbols_cover, and #717 holds unicode-width below 0.2.3 in Cargo.toml and .github/renovate.json until ratatui changes.

Acceptance criteria

• A ratatui-core release measures ｶﾞ as two cells with unicode-width 0.2.3 or later.
• The unicode-width bound in Cargo.toml and its rule in .github/renovate.json are gone, and the lock resolves unicode-width 0.2.3 or later with every test passing.

Notes

• As of 2026-10-28, ratatui-core 0.1.2 is the latest release, ratatui's main still adds the cell, and no ratatui issue or PR reports the double count. Its manifest accepts any unicode-width from 0.2.0.
• While the bound holds, ox's own layout code measures with unicode-width 0.2.2 directly, as wrap.rs does, and counts each mark as zero cells while ratatui paints one. Lifting the bound with a fixed ratatui brings both to one cell.
• ratatui's own cell_width tests, halfwidth_dakuten_alone and halfwidth_katakana_with_dakuten, fail against unicode-width 0.2.3, while its committed Cargo.lock stays on 0.2.2. Its weekly grouped Dependabot update waits 7 days after a release, so the first one to bring in 0.2.3, from about 2026-11-04, should surface the double count there. Whether to report it upstream before then is the maintainer's call.
• If a ratatui-core release starts requiring unicode-width 0.2.3 or later, it cannot resolve against the bound, so a failing Renovate Rust dependencies PR is also a cue to revisit this issue.
• fix(core): make Span::width count the cells that get rendered ratatui/ratatui#2790 (open) would make Span::width and Line::width count the cells ratatui renders. wrap.rs decides whether a line fits through Line::width, so that change would move this check onto ratatui's measure, while the grapheme loop that breaks long lines still calls unicode-width.