Stop the slugifier from deleting the letter D-stroke

ingestion/ingestion/segment/assembler.py

@@ -92,9 +92,21 @@ _Event = Union[Heading, _SectionEvent, _TextEvent]  # Heading == a title event
 
 
 def _slugify(text: str) -> str:
-    normalized = unicodedata.normalize("NFKD", text)
+    # `đ`/`Đ` (U+0111/U+0110) are standalone Vietnamese letters, not a base
+    # letter plus a combining mark, so NFKD leaves them whole and the ASCII
+    # encode below then discards them silently -- turning `ĐIỆN GIẢI` into
+    # `ien_giai`. Exactly three monographs contain `Đ` and all three carried a
+    # damaged drug_id because of this: GIẢI ĐỘC TỐ UỐN VÁN, KHÁNG ĐỘC TỐ BẠCH
+    # HẦU, and THUỐC UỐNG BÙ NƯỚC VÀ ĐIỆN GIẢI. Found via the adversarial eval
+    # suite, where both Oresol cases failed to resolve their monograph.
+    #
+    # Casefold before replacing so one pass covers both cases: `Đ` casefolds
+    # to `đ`. This matches what `entities/catalog.py::normalize_name` already
+    # does -- that function got it right and this one did not.
+    folded = text.casefold().replace("đ", "d")
+    normalized = unicodedata.normalize("NFKD", folded)
     ascii_text = normalized.encode("ascii", "ignore").decode("ascii")
-    return re.sub(r"[^a-z0-9]+", "_", ascii_text.lower()).strip("_")
+    return re.sub(r"[^a-z0-9]+", "_", ascii_text).strip("_")
 
 
 def _starts_its_visual_line(span: Span, previous: Span | None) -> bool: