Persian figure labels in numba.md are unshaped and reversed — matplotlib does no Arabic shaping or bidi

numba.md sets Persian text on matplotlib figures — ax.set_xlabel('تعداد نمونه‌ها') at line 820 and ax.set_ylabel(r'تخمین $\pi$') at 822, introduced by sync commit 7cd985b — and this repo carries no matplotlib font configuration and no font assets at all.

This is not the same problem as the Chinese editions, and it must not get the same fix. The glyphs do render: DejaVu Sans, matplotlib's default, covers all 29 codepoints involved. What fails is shaping. matplotlib performs no Arabic contextual shaping and no bidirectional reordering, so the label comes out as isolated letterforms in left-to-right order — legible as individual characters, wrong as Persian text. Swapping in a Persian font would not help, because the defect is in the text-layout stage rather than in glyph coverage.

Fix

The conventional approach is to reshape and reorder the string before it reaches matplotlib:

• arabic-reshaper to apply contextual forms
• python-bidi to apply the Unicode bidirectional algorithm

Both would need adding to environment.yml. Worth deciding whether to apply this per-label, via a small helper, or to avoid Persian in figure text altogether for now — the third option is legitimate and is what the rest of this edition currently does by default.

How it was found

A survey of CJK font handling across the translation program (project-translation reports/2026-07-24-cjk-font-rendering-review.md) checked the non-Latin editions for the same class of defect. This is the only instance in .fa; the .fr edition needs nothing, since DejaVu covers Latin with accents and French needs no shaping.

Filed for a decision rather than as an urgent break — one lecture, and the text is present rather than absent.

先從 numba.md 第 820-822 行和 environment.yml 開始，接著檢查其他翻譯版本如何處理非拉丁文字的圖表標籤。將提議的 shaping 相依項目與按標籤處理、使用 helper 以及避免文字這幾種方案進行比較；完成的標準是專案有一個經過測試並記錄在案的決定，並且如果採用 shaping，波斯語標籤能夠正確呈現。