fix: don't stub missing snapshot files with empty placeholders

download_snapshot wrote a 0-byte placeholder for any missing file,
including index.faiss. A 0-byte file is not a valid FAISS index and
crashes faiss.read_index instead of being treated as absent, breaking
the very first sync against a brand-new dataset repo. Missing files
are now left absent so load_index's existing FileNotFoundError
fallback (and sqlite3.connect's create-on-open behavior) apply as
intended.

sync/hf_dataset_store.py

@@ -16,7 +16,6 @@ def download_snapshot(repo_id: str, local_dir: str, token: str) -> dict:
         try:
             hf_hub_download(repo_id=repo_id, repo_type="dataset", filename=filename, local_dir=local_dir, token=token)
         except (FileNotFoundError, EntryNotFoundError):
-            open(os.path.join(local_dir, filename), "wb").close()
             continue
         if filename == "state.json":
             with open(os.path.join(local_dir, filename)) as f:

tests/sync/test_hf_dataset_store.py

@@ -1,7 +1,10 @@
 import json
 import os
 import httpx
+import pytest
 from huggingface_hub.errors import RemoteEntryNotFoundError
+from common.vector_index import load_index
+from common.db import init_db
 from sync.hf_dataset_store import download_snapshot, upload_snapshot
 
 
@@ -37,19 +40,28 @@ def test_download_snapshot_missing_state_returns_empty_state(tmp_path, monkeypat
     monkeypatch.setattr("sync.hf_dataset_store.hf_hub_download", fake_hf_hub_download)
     state = download_snapshot("org/repo", str(tmp_path), token="tok")
     assert state == {"papers": {}, "last_synced_at": None}
+    assert not os.path.exists(tmp_path / "state.json")
 
 def test_download_snapshot_handles_empty_remote_dataset_repo(tmp_path, monkeypatch):
     """Reproduces a real first-time sync against a brand-new, empty dataset repo:
     huggingface_hub raises RemoteEntryNotFoundError (an HTTP 404), not a plain
-    FileNotFoundError, when a file doesn't exist yet on the Hub."""
+    FileNotFoundError, when a file doesn't exist yet on the Hub. Missing files
+    must be left absent (not stubbed with an empty placeholder), since a 0-byte
+    file is not a valid "missing" signal to faiss.read_index / sqlite3.connect."""
     def fake_hf_hub_download(repo_id, repo_type, filename, local_dir, token):
         raise _remote_entry_not_found()
 
     monkeypatch.setattr("sync.hf_dataset_store.hf_hub_download", fake_hf_hub_download)
     state = download_snapshot("org/repo", str(tmp_path), token="tok")
     assert state == {"papers": {}, "last_synced_at": None}
-    assert os.path.exists(tmp_path / "index.faiss")
-    assert os.path.exists(tmp_path / "papers.db")
+    assert not os.path.exists(tmp_path / "index.faiss")
+    assert not os.path.exists(tmp_path / "papers.db")
+
+    # Downstream loaders must fall back cleanly, exactly as run_sync relies on.
+    with pytest.raises(FileNotFoundError):
+        load_index(str(tmp_path / "index.faiss"))
+    conn = init_db(str(tmp_path / "papers.db"))
+    assert conn.execute("SELECT COUNT(*) FROM papers").fetchone()[0] == 0
 
 def test_upload_snapshot_uploads_folder_as_single_atomic_commit(tmp_path, monkeypatch):
     for name in ["index.faiss", "papers.db", "state.json"]: