fix: classify choose() emoji keys by unicode properties, not length+ascii

_looks_unicode_emoji used 'len<=4 and not isascii', so a short non-ascii text label
(Sí/да/はい/café/确定) was misread as a unicode emoji and sent as a labelless component
emoji, which Discord 400s. It now treats a key as an emoji only when every codepoint is
emoji-composition material (So/Sk symbols, ZWJ, variation selectors, regional indicators,
keycap combiner) with at least one pictographic symbol - so a compound emoji (ZWJ family,
flag, skin-tone, keycap) stays one emoji while an i18n label is a label. The terminal
Discord 400 remains a documented live-gap (needs a gateway).

Signed-off-by: disqualifier <dev@disqualifier.me>

src/dpy_commons/prompts.py

@@ -10,6 +10,7 @@ treating a timeout as a normal None return.
 from __future__ import annotations
 
 import logging
+import unicodedata
 from typing import Any, Optional, Union
 
 import discord
@@ -227,6 +228,43 @@ def _split_key(key: Any) -> "tuple[Optional[str], Optional[Union[str, discord.Pa
     return str(key), None
 
 
+def _is_emoji_codepoint(ch: str) -> bool:
+    """whether a single character is emoji-composition material
+
+    covers the pictographic symbols themselves (category So), skin-tone modifiers (Sk),
+    and the glue that binds a compound emoji into one grapheme: ZWJ (U+200D), variation
+    selectors (U+FE00-FE0F), regional-indicator letters (flags, U+1F1E6-1F1FF), and the
+    keycap combiner (U+20E3). a plain letter/digit/space is none of these.
+    """
+    code = ord(ch)
+    if ch == "‍" or ch == "⃣":
+        return True
+    if 0xFE00 <= code <= 0xFE0F:
+        return True
+    if 0x1F1E6 <= code <= 0x1F1FF:
+        return True
+    return unicodedata.category(ch) in ("So", "Sk")
+
+
 def _looks_unicode_emoji(value: str) -> bool:
-    """heuristic: a short non-ascii token is treated as a unicode emoji key"""
+    """whether value is a unicode emoji key rather than a text label
-    return bool(value) and len(value) <= 4 and not value.isascii()
+
+    true only when EVERY codepoint is emoji-composition material (see _is_emoji_codepoint)
+    and at least one is a pictographic symbol - so a compound emoji (ZWJ family, flag,
+    skin-tone) classifies as one emoji, while a short non-ascii text label (Sí/да/はい/確定/
+    café) is a label because its letters are not emoji codepoints. does not cap length:
+    a long emoji-only run is still an emoji, a run mixing letters and emoji is a label.
+    """
+    if not value:
+        return False
+    # a keycap sequence is [0-9#*] + optional VS16 + U+20E3 - its base is an ascii digit/
+    # symbol (not So), so treat the whole run as an emoji when the keycap combiner is present
+    if value.endswith("⃣") and all(c in "0123456789#*️⃣" for c in value):
+        return True
+    has_symbol = False
+    for ch in value:
+        if not _is_emoji_codepoint(ch):
+            return False
+        if unicodedata.category(ch) == "So":
+            has_symbol = True
+    return has_symbol