What this PR does / why we need it:

Online serving drops python mode on demand feature views when full_feature_names=True.

store.get_online_features(
    entity_rows=[{"driver_id": 1001}],
    features=["driver_hourly_stats:conv_rate", "python_view:conv_rate_plus_acc"],
    full_feature_names=True,
)
# -> {"driver_id": [...], "driver_hourly_stats__conv_rate": [...]}
#    python_view__conv_rate_plus_acc is missing, no error raised

In _augment_response_with_on_demand_transforms, a full-name request looks for "<view>__<feature>". The pandas path gets that name because transform_arrow renames its outputs. The python path uses transform_dict, which returns the UDF's short output names. Nothing matches, so selected_subset is empty and the feature disappears from the response. Batch and singleton=True python ODFVs are both affected. A model client that asks for full names gets a KeyError or a silently missing input, and the same request in pandas mode works.

The fix adds a full_feature_names parameter to OnDemandFeatureView.transform_dict, as suggested in review. A new _postprocess_feature_dict renames outputs the same way _postprocess_arrow_table does, so transform_dict, transform_arrow and transform_ibis all name their outputs consistently for any caller. The online path passes full_feature_names through. The declared schema type is also looked up under the full name. Without that, full-name values fell back to type inference instead of the declared dtype.

Which issue(s) this PR fixes:

No existing issue. I searched open issues and PRs for full_feature_names with python mode or ODFVs and didn't find a match.

This PR also adds the alias + python mode + full_feature_names coverage requested by @ntkathole on #6901.

Checks

• I've made sure the tests are passing.
• My commits are signed off (git commit -s)
• My PR title follows conventional commits format

Testing Strategy

• Unit tests
• Integration tests
• Manual tests
• Testing is not required for this change

Branch rebased on master @ 9d42729 (includes #6901).

• test_python_views_with_full_feature_names (test_on_demand_python_transformation.py): requests a pandas ODFV, a batch python ODFV and a singleton python ODFV with full_feature_names=True. On the original base (3667772), only pandas_view__conv_rate_plus_acc_pandas came back and both python outputs were missing. Passes with this change. The aliased test below shows the same python mode gap on current master @ 9d42729.
• test_feature_service_serves_aliased_odfv_features[pandas|python|python_singleton] (test_on_demand_pandas_transformation.py): an ODFV added to a feature service with with_name("scaled_alias"), served with default and full names, so this covers alias + python mode + full_feature_names together.
  • On master (has fix: Serve aliased on demand feature views in feature services #6901, not this fix): [pandas] passes; [python] and [python_singleton] fail with Right contains one more item: 'scaled_alias__conv_rate_x10'.
  • With fix: Serve aliased on demand feature views in feature services #6901's alias lookup reverted on this branch: all three fail with KeyError: 'conv_rate_x10'.
  • With this change: all three pass.
• test_transform_dict_full_feature_names[batch|singleton]: calls transform_dict directly and checks short names by default, plus_one__val_plus_one with full_feature_names=True, and alias__val_plus_one after with_name("alias"). It fails against the first revision of this PR (TypeError: transform_dict() got an unexpected keyword argument 'full_feature_names') and passes now.
• Across the ODFV, transformation, feature service, online response, versioning, utils and feature store unit files, this branch gives 184 passed and master gives 180. Both have the same 7 failures and 7 collection errors locally: missing torch, the TestOnDemandTransformationsWithWrites UDF globals NameError tracked in On-demand feature view UDFs lose their module globals when rebuilt from the registry, so serving fails with NameError #6815, two serve_registry tests, and two modules that need deps not installed here.
• ruff check, ruff format --check and mypy feast/on_demand_feature_view.py feast/utils.py are clean.

Misc

This change was prepared with an AI coding assistant, and the tests and checks above were run locally.