fix(glossy): use codepoints (not bytes!) in article code

Fixes [#18](https://github.com/swaits/typst-collection/issues/18)

glossy/src/gloss.typ

@@ -605,7 +605,7 @@
         // 1. create array of tuples with (lower [if ignore-case], entry)
         .map(e => { if ignore-case { (lower(e.short), e) } else { (e.short, e) } })
         // 2. sort the tuples (by first element then second)
-        .sorted(key: t => t.first())
+        .sorted() // NOTE: sorted() is NOT language-aware
         // 3. strip away the tuple's first element, leaving an array of entries
         .map(t => t.last())
 

glossy/src/utils.typ

@@ -203,7 +203,7 @@
 
   if is_acronym {
     // Use acronym rules
-    let first_char = upper_word.slice(0,1)
+    let first_char = upper_word.codepoints().slice(0,1).first()
     if acronym_vowels.contains(first_char) {
       "an"
     } else {
@@ -211,7 +211,7 @@
     }
   } else {
     // Non-acronym rules
-    let first_char = lower_word.slice(0,1)
+    let first_char = lower_word.codepoints().slice(0,1).first()
 
     // Special cases for words matching a set of known 'silent h' prefixes
     let silent_h_words = (