Signed-off-by: Nick Quinn <nicholas_quinn@apple.com>

This commit fixes three critical issues to properly support OnDemandFeatureViews without transformations:

1. on_demand_feature_view.py:
   - Allow feature_transformation to be None (passthrough mode)
   - Fix __init__ to not raise ValueError when both feature_transformation and udf are None
   - Only create FeatureTransformationProto when transformation exists
   - Add null checks before accessing feature_transformation

2. utils.py:
   - Fix passthrough ODFVs being skipped (continue at line 679-680)
   - Extract features from initial response for passthrough ODFVs
   - Unify transformed_features variable across all code paths
   - Fix aggregation results assignment

3. Java interop:
   - Ensures hasFeatureTransformation() in Java correctly returns false for passthrough ODFVs
   - Prevents unnecessary transformation service calls

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Nick Quinn <nicholas_quinn@apple.com>

The infer_features() method at line 979 was calling
self.feature_transformation.infer_features() without checking if
feature_transformation is None. This caused failures during 'feast apply'
for passthrough ODFVs (those without transformations).

Changes:
- Add early return for passthrough ODFVs if features are already defined
- Require explicit schema for passthrough ODFVs (cannot be inferred)
- Raise clear error message if passthrough ODFV has no schema
- Maintain existing behavior for ODFVs with transformations

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Nick Quinn <nicholas_quinn@apple.com>

Signed-off-by: Nick Quinn <nicholas_quinn@apple.com>

Signed-off-by: Nick Quinn <nicholas_quinn@apple.com>