[dvd233]

Checklist

• Regression tests are included
• Commit message follows commit guidelines
• npm test passes without local adaptation (Windows validation note below; upstream CI pending)

Description of change

Fixes #7292.

Replaying a categorical brush:highlight selection currently reads only its first two values. A single category such as ['H'] produces invalid mask coordinates, while ['H', 'I', 'J'] highlights only H and I. Brush events themselves emit the complete category list, so forwarding event.data.selection can lose the original selection.

Use the first and last domain values in abstractOf, preserving the existing band-width adjustment. This supports single and complete multi-category selections while retaining two-endpoint and continuous selections.

Add 12 scale tests and 7 Chart/SVG integration tests covering emitted-selection replay, finite mask geometry, point scales, and moving/resizing a restored brush. The integration tests dispatch scenegraph drag events; they are not browser mouse E2E tests.

Validation:

• The unchanged regression tests produce 13 failures and 6 passes on the original source, and all 19 pass with the fix.
• Full suite: 962 passed, 25 existing skips; all 968 existing test statuses match the baseline (943 passed, 25 skipped). No golden changes.
• Full ESLint and Prettier checks, CJS/ESM/UMD builds and bundle-size checks pass. ESLint reports 24 warnings in unchanged files.

Windows validation note: both full runs use SVGO's supported js2svg.eol: 'lf' option through a local runtime transform to match the repository's LF SVG goldens. This does not change assertions, goldens, test selection or coverage, and is not included in the PR.

Prepared with AI assistance; the source change and regression evidence were reviewed before submission.

[hustcc]

@dvd233 已经复现问题，并测试这个分支解决了这个问题吗？

[dvd233]

是的，提交前已用相同回归测试对比原源码和本分支：原来单个分类 ['H'] 会错误高亮全部 5 项，['H', 'I', 'J'] 只高亮 H/I；19 个新增用例在原源码上 13 失败、6 通过，本分支全部 19 通过。其中 7 个是实际 Chart/SVG 场景，覆盖选区事件回放、有限坐标以及恢复后移动/缩放；它们派发 scenegraph 拖动事件，不是浏览器鼠标 E2E。完整测试当时为 962 通过、25 个原有跳过，Windows 的本地 LF 序列化适配和验证范围已在 PR 正文注明。当前 head 与这份已验证提交一致。