remove weights_are_all_positive_ from TreeEnsemble

[xadupre]

Description

Remove weights_are_all_positive_. This attribute was part of the code because the post transform was originally shared with SVM. This is no longer the case.

Motivation and Context

Fixes #27533.

[github-actions]

You can commit the suggested changes from lintrunner.

[Copilot]

Pull request overview

This PR fixes a TreeEnsembleClassifier binary edge case where post_transform=LOGISTIC could incorrectly skip applying sigmoid when the model is a degenerate “leaf-only” tree with all non-negative weights, by removing the weights_are_all_positive_ heuristic from the TreeEnsemble implementation and adding a Python regression test.

Changes:

• Remove weights_are_all_positive_ from TreeEnsembleCommonClassifier initialization and from TreeAggregatorClassifier.
• Update binary-case scoring to always use the “logit-style” path (add_second_class = 2/3) so LOGISTIC is applied consistently.
• Add a Python unit test covering split-tree, leaf-only, and leaf-only-with-negative-weight scenarios for post_transform=LOGISTIC.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
onnxruntime/test/python/onnxruntime_test_python.py	Adds a regression test for issue #27533 validating sigmoid application in binary TreeEnsembleClassifier LOGISTIC mode, including leaf-only trees.
onnxruntime/core/providers/cpu/ml/tree_ensemble_common.h	Removes weights_are_all_positive_ tracking and simplifies binary-case detection via class-id set construction.
onnxruntime/core/providers/cpu/ml/tree_ensemble_aggregator.h	Removes the all-positive-weights branch and forces binary-case scoring to route through add_second_class 2/3 so LOGISTIC applies sigmoid.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

You can commit the suggested changes from lintrunner.