fix: Handle empty label_enumeration metadata and correct query dtypes in filtered Vamana tests

  Fixes 6 test failures where empty filter metadata strings caused JSON decode errors,
  and query vectors were sliced from float64 arrays instead of the float32 vectors array.

Author: brooksomics

apis/python/src/tiledb/vector_search/vamana_index.py

@@ -254,7 +254,7 @@ def query_internal(
         if where is not None:
             # Get label enumeration from metadata
             label_enum_str = self.group.meta.get("label_enumeration", None)
-            if label_enum_str is None:
+            if not label_enum_str:
                 raise ValueError(
                     "Cannot use 'where' parameter: index does not have filter metadata. "
                     "This index was not created with filter support."

apis/python/test/test_filtered_vamana.py

@@ -141,7 +141,7 @@ def test_filtered_query_equality(tmp_path):
     index = VamanaIndex(uri=uri)
 
     # Query near cluster A with filter for dataset_A
-    query = vectors_cluster_a[0:1]  # Use first vector from cluster A
+    query = vectors[0:1]  # Use first vector from cluster A (dataset_A)
     distances, ids = index.query(query, k=k, where="label == 'dataset_A'")
 
     # Verify all results are from dataset_A (IDs 0-249)
@@ -227,7 +227,7 @@ def test_filtered_query_in_clause(tmp_path):
     index = VamanaIndex(uri=uri)
 
     # Query with IN clause for datasets 1 and 3
-    query = vectors_a[0:1]
+    query = vectors[0:1]  # Use first vector from cluster A (soma_dataset_1)
     distances, ids = index.query(
         query, k=k, where="label IN ('soma_dataset_1', 'soma_dataset_3')"
     )