refactor(attr-inference+data-recon): migrate to new attack architecture

[asim29]

Context

Seventh in the refactor/risk-modules stack. Stacked on #79 (discriminatory behavior). Final risk-vertical PR.

Changes

amulet/attribute_inference/attacks/duddu_cikm_2022.py

• DudduCIKM2022 subclasses AttributeInferenceAttack(ABC) and delegates target_model, x_train_adv, x_test, z_train_adv, device to super().__init__().
• attack() return type narrowed to dict[int, dict[str, np.ndarray]].

amulet/attribute_inference/metrics/attack_accuracy.py

• Docstring converted to Google style; :class: RST markup removed.

amulet/data_reconstruction/attacks/fredrikson_ccs_2015.py

• FredriksonCCS2015 subclasses DataReconstructionAttack(ABC) and delegates target_model, input_size, output_size, device to super().__init__().

amulet/data_reconstruction/metrics/similarity_evaluation.py

• Module-level __figure_mse (mangled name, invisible outside module) replaced with a module-private _figure_mse helper. Docstring converted to Google style.

examples/attack_pipelines/run_attribute_inference.py / run_data_recon.py

• Updated to use load_or_train instead of manual save/load blocks.

tests/integration/test_other_risks.py (new)

• Smoke tests for DudduCIKM2022.attack(), FredriksonCCS2015.attack(), and SuriEvans2022.attack().

tests/integration/test_celeba_datasets.py, test_lfw_datasets.py, test_utkface_datasets.py (new)

• Integration tests for the CelebA, LFW, and UTKFace dataset loaders (marked @pytest.mark.integration; require network/disk access).

tests/unit/test_metrics_attribute.py (new)

• Unit tests for evaluate_attribute_inference covering perfect-prediction identity and output shape.

Test plan

• uv run pytest tests/integration/test_other_risks.py -v -m integration
• uv run pytest tests/unit/test_metrics_attribute.py -v
• uv run pre-commit run --all-files
• CI passes on this branch