fix: deterministic pagination order for derived conclusions

created_at is the transaction timestamp, so conclusions derived in the
same batch share it; add an id tie-breaker (mirroring the created_at
direction) so /derived pagination can't skip or duplicate rows across
page boundaries. Matches the existing precedent in
query_documents_most_derived.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/crud/document.py

@@ -1394,9 +1394,13 @@ def get_child_observations(
     if observed:
         stmt = stmt.where(models.Document.observed == observed)
 
+    # created_at is the transaction timestamp, so documents created in the
+    # same batch share it -- id keeps pagination deterministic.
     if reverse:
-        stmt = stmt.order_by(models.Document.created_at.asc())
+        stmt = stmt.order_by(models.Document.created_at.asc(), models.Document.id.asc())
     else:
-        stmt = stmt.order_by(models.Document.created_at.desc())
+        stmt = stmt.order_by(
+            models.Document.created_at.desc(), models.Document.id.desc()
+        )
 
     return stmt