fix: exclude exponential loss for GradientBoostingClassifier on multiclass targets

[kimusaku]

Problem

Closes #64.

sklearn’s ExponentialLoss only supports binary classification. When hyperparameter tuning was enabled with GradientBoostingClassifier on a multiclass target, optuna would occasionally sample loss='exponential', raising:

ValueError: ExponentialLoss requires 2 classes; got 3 class(es).

The same bug also listed 'deviance' as a loss candidate — that alias was removed in sklearn 1.3 and would raise ValueError on any modern install.

Changes

sapientml_core/templates/model_templates/hyperparameters.py.jinja

• Gate 'exponential' behind {% if not is_multiclass %} — binary-only, never emitted for multiclass targets.
• Remove 'deviance' entirely (removed in sklearn 1.3).

sapientml_core/adaptation/generation/pipeline_template.py

• Forward pipeline.task.is_multiclass into the hyperparameters template render context so the Jinja guard has the information it needs.

tests/sapientml/test_hyperparameters_template.py

• Update _render helper to accept an optional is_multiclass parameter (default False, preserving all existing tests).
• Add four new regression tests:
  1. test_gradient_boosting_classifier_multiclass_excludes_exponential — verifies 'exponential' never appears when is_multiclass=True.
  2. test_gradient_boosting_classifier_binary_includes_exponential — verifies 'exponential' is available for binary classification.
  3. test_gradient_boosting_classifier_always_excludes_deviance — verifies 'deviance' never appears in any rendering.
  4. test_gradient_boosting_multiclass_tuning_completes_without_error — end-to-end optuna study on a synthetic 3-class dataset that must complete all trials without ValueError.

Test results

All 38 tests pass (pytest tests/sapientml/test_hyperparameters_template.py).