Write Roman front matter and body page labels with outlines

[chroming]

Closes #60

When the first body page is physical PDF page 31, exporting a directory now writes both the chapter outline targeting page 31 and PDF page labels that display that page as 1 (xxx on page 30, 1 on page 31). Existing page labels, including custom prefixes and multiple ranges, are preserved by default.

The GUI adds a reader-page-number option beside page offset, with an inferred or manually selected body start. The CLI exposes the same behavior through --page-labels roman-body and optional --body-start-page. Export validates page targets and labels, checks that the source did not change during writing, and replaces an existing output only after a verified temporary PDF is ready. The GUI confirms replacement, routes both typed paths and the Open dialog through the same unsaved-draft decision, and reports the final export status.

Validation:

• 55 tests passed; the one deselected tests/test_convert.py::test_convert_dir_text also fails on unmodified origin/master.
• Relevant PDF and GUI regression tests passed with the declared pypdf 3.17.4 dependency (10 passed).
• CLI and macOS Cocoa GUI exports were reopened and checked: physical page 31 has label 1, and the chapter outline targets physical page 31.
• Packaged-app and third-party PDF-reader checks were not run.

[github-actions]

✅ OpenCodeReview: Review partially complete: 0 finding(s); 6 of 7 selected item(s) failed.

[chroming]

已在 a98fd91 处理审查意见：修复无效页码编辑导致切换 PDF 异常、导出权限回归和 CLI 越界报错；合并重复的页数读取、补充错误日志和校验；将新增 GUI 文案接入 Qt 翻译，并明确负页差时的手动操作。相关回归测试已补齐。当前测试为 60 passed、1 个既有失败（tests/test_convert.py::test_convert_dir_text，相关代码相对 origin/master 未改动）。

Windows 句柄那条未改：当前使用的 pypdf 3.17.4 在 PdfReader(path) 内通过 with open 读取到 BytesIO，随后立即关闭文件句柄；这里不会因 reader 持有临时文件句柄而阻止 os.replace。