Arabic TTS reads no digits at all, not even Arabic-Indic digits

Problem
When using Arabic TTS (ar-MSA), no digit system works — not ASCII (0-9), not Arabic-Indic (٠-٩). All digits are silently dropped.
Example
Arabic-Indic digits (٠-٩) — native Arabic digits
./magpie-cli say --model ... --lang ar-MSA
--text "وزارة العمل تعلن عن ٨٥٥ فرصة عمل حتى سن ٥٠ عاما بـ ٣ مدن"
Audio Output: "وزارة العمل تعلن عن فرصة عمل حتى سن عاما بـ مدن" (ALL numbers missing)

[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '٨' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '٥' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '٥' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '٥' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '٠' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char 'ـ' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '٣' skipped

ASCII digits (0-9)
./magpie-cli say --model ... --lang ar-MSA
--text "وزارة العمل تعلن عن 855 فرصة عمل"
Audio Output: "وزارة العمل تعلن عن فرصة عمل" (numbers missing)
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '8' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '5' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '5' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '5' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '0' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char 'ـ' skipped
[magpie] tokenizer arabic_MSA_chartokenizer: unknown char '3' skipped

Possible Root Cause [suggested by opencode/nemotron-3-ultra]
The ArabicCharsTokenizer (charset v1) uses a fixed character vocabulary loaded from the GGUF model. This vocabulary appears to not include any digit characters at all — not even Arabic-Indic digits (٠-٩, U+0660–U+0669).
The tokenizer does per-character lookup (st.in_vocab(c) at tokenizer.cpp:580) and drops any character not in the vocabulary with a warning log.
Expected Behavior
Arabic TTS should read numbers written in Arabic-Indic digits (٠-٩) at minimum, since that's the standard digit system for Arabic text.
Affected Languages

• ar-MSA (Modern Standard Arabic)

it uses the same ArabicCharsTokenizer with "fixed charset v1" (per architecture doc).
Current Workaround
None — there is no way to synthesize spoken numbers in Arabic with the current model/tokenizer. Numbers must be written out in words (e.g., "ثمانمائة وخمسة وخمسون" instead of "855" or "٨٥٥").

Start with ArabicCharsTokenizer and the per-character lookup at tokenizer.cpp:580, then consult the architecture documentation for the fixed charset v1 model vocabulary. Reproduce the ar-MSA examples with both Arabic-Indic and ASCII digits and trace the unknown-character warnings. Done means digits are no longer dropped and Arabic TTS synthesizes the numbers in those inputs.