Fix CI: make tests hermetic and harden hybrid_rank against attr-less chunks

CI installs fastembed (requirements-dev -> requirements), which activated the
dense retrieval path during tests and surfaced two issues:
- hybrid_rank read chunk.embedding directly, crashing on the SimpleNamespace
test stand-ins (and any chunk-like object without the attribute). Use
getattr(c, embedding, None) so such a chunk simply stays lexical.
- The suite would download the ~130 MB embedding model on every run (slow and
flaky). conftest now disables embeddings, so tests are hermetic and
deterministic; the TF-IDF+vector fusion is already unit-tested with mocked
vectors in test_embeddings.py.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/app/retrieval.py

@@ -110,7 +110,9 @@ def hybrid_rank(
         return tfidf_scored[:top_k]
 
     # Semantic ranking over every chunk that has a stored vector (one matmul).
-    pairs = [(c, embeddings.to_vector(c.embedding)) for c in chunks]
+    # getattr keeps this robust to any chunk-like object without an embedding
+    # attribute — such a chunk simply contributes no vector and stays lexical.
+    pairs = [(c, embeddings.to_vector(getattr(c, "embedding", None))) for c in chunks]
     pairs = [(c, v) for c, v in pairs if v is not None and v.shape == qvec.shape]
     if not pairs:
         return tfidf_scored[:top_k]

backend/tests/conftest.py

@@ -14,9 +14,16 @@ from fastapi.testclient import TestClient
 from sqlalchemy import create_engine
 from sqlalchemy.orm import sessionmaker
 
+from app import config
 from app.database import Base, get_db
 from app.main import app
 
+# Keep the suite hermetic and deterministic: the optional dense-embedding model
+# (fastembed) is a ~130 MB network download that would make CI slow and flaky, so
+# tests run on pure TF-IDF. The hybrid TF-IDF+vector fusion itself is covered
+# deterministically with mocked vectors in test_embeddings.py.
+config.EMBED_ENABLED = False
+
 
 @pytest.fixture()
 def _engine():