More robust performance test

[bejaeger]

Issue

Previously, the test was prone to statistical fluctuations because we were fitting a very small dataset.
This creates a dataset specific to the test and uses a few more samples to make the test more robust.

[gemini-code-assist]

Code Review

This pull request aims to make a test more robust by replacing a 3-class classification dataset with an easier 2-class dataset, reducing the likelihood of failures due to statistical fluctuations. The change is well-implemented. My feedback includes a suggestion to strengthen the performance assertions in the test to better leverage the simpler dataset and improve the test's ability to detect regressions.

[Copilot]

Pull Request Overview

This PR modifies the test_predict_logits_and_consistency test to use inline dataset generation instead of the shared X_y fixture. The test now creates its own classification dataset with specific parameters (60 samples, 2 classes, 3 features) rather than using the module-level fixture that generates a different dataset (60 samples, 3 classes, 5 features).

Key Changes:

• Removed the X_y fixture parameter from the test function signature
• Added inline sklearn.datasets.make_classification() call with custom parameters (2 classes vs 3, 3 features vs 5)

---

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

[brendan-priorlabs]

Thanks, @bejaeger!