feat(editor): inline writing review / 文档即时审校 (v0.38.1)

[zeweihan]

中文

输入正文后自动运行本地规则检查，在编辑器显示当前段落提示和问题清单。编号、待填内容、字形、算术、统一社会信用代码和项目资料引用使用本地检查；只有点击“深入审校”才调用 AI，外部库查询仍需用户明确操作。

移除工具栏“解析”。依据窗格保留全文在线核验与旧结果，打开窗格不消费额度；用户点击在线核验后先保存当前文档。提示定位绑定会话、文档版本、段落与逐字原文，过期结果拒绝操作，采用建议保留修订并支持一步撤销。范围明确为正文段落，超限提示截断。版本 0.38.1。

验证：后端全套 3494 项（10 跳过）；最后规则边界修改后针对性 72 项；前端写作/审校/依据回归 194 项，新增浏览器计时器及视口回归另验；完整 LOWA 回归 547/547；真实 LOWA 定位/修订/撤销、中文 IME、手动 AI 请求、滚动与过期隐藏通过；真实 Electron 中文补全→继续输入待填写→自动即时提示→保存下载通过，自动 AI/外查为零。150 页大文档 3 轮对照：读取 15ms、查找 3.10s、格式化 18.86s；导出 10.78s 超出原有 10s 阈值，发布前基线同样为 10.71s（各 3 轮），未修改阈值；两版导出后均无保存循环。

整应用中英文旅程 97 项通过；AI 消费、多用户及资源包环境相关旅程按套件开关跳过。

开发看板：zeweihan/dev-board#547。设计、实现计划及 Claude 领域契约随 PR 更新。

English

Review live Writer body text after typing, with local rule findings beside the current paragraph and in an issue list. AI runs only after an explicit Deep review click; external source lookups remain explicit. Remove the Parse toolbar button while retaining saved results and explicitly requested full online verification, with a same-document save barrier.

Session/revision/paragraph/quote guards reject stale locations and edits. Accepted suggestions retain tracked changes and one-step undo. Body-only scope and truncation are disclosed. Version 0.38.1.

Whole-app Chinese/English journeys: 97 passed; paid AI, multi-user and resource-pack environment journeys skipped via suite switches.

Validation: full backend suite (3494 tests, 10 skipped), 72 targeted checks after final rule changes, frontend review/writing regressions, 547 passing real LOWA checks, and real Electron Chinese completion/live review/save/download. Typing issued no AI or external lookup. Three-run 150-page comparison: export remains above the existing 10s threshold (10.78s vs baseline 10.71s); the threshold is unchanged. Both versions have zero post-export modified events.

长文档行内定位补测 / Large-document inline context: paragraph 919 in a 150-page fixture; cold index 2.24s, subsequent reads 58–113ms. Checks remain debounced and IME-aware.