fix: correct inverted default_value logic in _construct_random_input

When singleton=True, the fallback should be scalar None, not [None].
When singleton=False, the fallback should be [None], not None.

The condition was inverted, causing incorrect UDF schema inference
for feature types not present in sample_values.

Fixes #6008

Signed-off-by: Br1an67 <932039080@qq.com>

Author: Br1an67

sdk/python/feast/on_demand_feature_view.py

@@ -1011,7 +1011,7 @@ def _construct_random_input(
             sample_values = {k: v[0] for k, v in sample_values.items()}
 
         # Default value for missing types
-        default_value = None if not singleton else [None]
+        default_value = [None] if not singleton else None
 
         feature_dict = {}