Neutral runs in Arabic text — where a mark is actually required
===============================================================
Each case below is given twice: WITHOUT the mark (how the string usually arrives from a
translator) and WITH it (what renders correctly). The two lines differ by exactly one invisible
character, so a diff tool that ignores them will show them as identical.

1. A trailing Latin identifier followed by Arabic punctuation
   without: تم رفع الملف report-2026.pdf.
   with:    تم رفع الملف ⁨report-2026.pdf⁩.
   why: the full stop is neutral and sits between an LTR run and the RTL paragraph, so without the
        isolate it can attach to the wrong side and render at the far left of the line.

2. A phone number inside an Arabic sentence
   without: اتصل على 555-0143 للحصول على المساعدة.
   with:    اتصل على ؜555-0143؜ للحصول على المساعدة.
   why: the hyphen between two European digit runs is neutral; ALM pins the whole number so the
        segments do not swap and render as 0143-555.

3. A version range
   without: الإصدارات 2.4 - 3.1 مدعومة.
   with:    الإصدارات ⁦2.4 - 3.1⁩ مدعومة.
   why: the spaces and the hyphen are all neutral, and the two numbers reorder without the isolate.

4. A file path with a mixed-script component
   without: احفظ في C:\Users\رنا\Documents
   with:    احفظ في ⁦C:\Users\⁨رنا⁩\Documents⁩
   why: nested isolates — an RTL name inside an LTR path inside an RTL sentence. FSI is used for
        the name because its direction comes from the data, not from the template.

5. A percentage and a currency amount
   without: تم استخدام 87% من 12,345.60 EGP
   with:    تم استخدام ؜87%؜ من ؜12,345.60 EGP؜
   why: the percent sign and the comma are neutral; ALM keeps each measurement together.

6. A bracketed LTR insertion, showing bracket MIRRORING
   without: راجع الدليل (Novus File Manager) قبل المتابعة.
   with:    راجع الدليل (⁨Novus File Manager⁩) قبل المتابعة.
   why: in an RTL paragraph the glyph U+0028 LEFT PARENTHESIS is drawn as ")" — mirroring is a
        RENDERING property, so the stored code point never changes and a byte test must not
        expect it to.

Assert on: (a) the "with" strings still contain their marks after a JSON/PO/XLIFF round trip,
(b) a bidi implementation produces the same visual order for the "with" strings under both
paragraph directions, and (c) stripping the marks turns each "with" line into its "without" twin.