Fix 15+ Unicode bypass strategies in chat filter

[Copilot]

Chat filter failed to catch obfuscated profanity using Unicode confusables, Chinese characters, zero-width characters, and various encoding tricks (fullwidth: ｎｉｇｇｅｒ, Chinese: 黑鬼, mixed scripts: ᛔitϹh, subscripts: ₛₕᵢₜ, etc.).

Changes

Extended Unicode normalization

• Added 100+ character mappings to UNICODE_CONFUSABLES: fullwidth Latin, mathematical alphanumerics (bold/italic), small caps, sub/superscripts, upside-down chars, additional Greek/Cyrillic/runic confusables
• Strip zero-width chars (U+200B-D, U+FEFF) and combining diacriticals (U+0300-036F, U+20D0-20FF) in normalizeObfuscatedSegment()

Chinese slur detection

• Added CHINESE_SLUR_MAP with severity annotations (黑鬼 → nigger, 傻逼 → fuck, etc.)
• Optimized with CJK regex pre-check before map iteration

Reversed text detection

• Check normalized text in reverse in segmentHasBadWord() to catch backwards obfuscation

Compound word detection

• Extended SEVERE_SUBSTRING_ROOTS to catch embedded profanity (jollycock → cock)

CJK character recognition

• Updated isWordLikeCharacter() to treat CJK ideographs (U+4E00-9FFF) as word chars for proper censoring

// Before: ｎｉｇｇｅｒ bypassed
// After: caught via fullwidth → ASCII normalization

// Before: 黑鬼 bypassed  
// After: caught via CHINESE_SLUR_MAP

// Before: ʇᴉɥs bypassed
// After: caught via upside-down normalization + reversal

Added 17 bypass strategy tests. All 40 tests passing.

Original prompt

Fix at least 15 ways/ strategies to bypass the cht filtering including things like CanadaBinny: ｎｉｇｇｅｒ, just now 🤖 and things like ⓝⓘⓖⓖⓔⓡ this was caught
Ꭰick, this was caught
@ss was caught

黑鬼, bypassed
ᛔitϹh? Ϲunt? bypassed
Shi十 bypassed
jollycock bypassed

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[BaconCat1]

@copilot just make it ban all the special characters.

[Copilot]

@copilot just make it ban all the special characters.

Banning all special characters would block legitimate messages with punctuation, emojis, and non-English text (like "Hello! 😊" or "¿Cómo estás?").

The current approach normalizes special characters to their ASCII equivalents (ｎ→n, 𝐧→n, ⓝ→n) before checking, so we can still detect "ｎｉｇｇｅｒ" as "nigger" while allowing normal special character usage.

If you want stricter filtering, I can:

1. Block messages with excessive special characters (>50% of message)
2. Add a whitelist of allowed special characters (punctuation, common emojis)
3. Block specific Unicode ranges entirely

Which approach would you prefer?