fix: collection field extraction and add regression tests for sort-key and nested layout

src/buddhagpt/corpus.py

@@ -23,9 +23,11 @@ def load_bilara_suttas(root: Path) -> Iterator[dict]:
         segments = json.loads(f.read_text())
         uid = f.name.split("_")[0]
         title = next(iter(segments.values()), uid)
+        # Derive collection from first directory under sutta/ (e.g., an, dn, kn, mn, sn)
+        collection = f.relative_to(base).parts[0]
         yield {
             "uid": uid,
             "title": title.strip(),
             "text": segments_to_text(segments),
-            "collection": f.parent.name,
+            "collection": collection,
         }