PR Summary

This PR introduces several functional improvements focused on enhancing unit tests and output handling for time series data validation and metrics reporting. The changes include:

1. In the ADF (Augmented Dickey-Fuller) tests, the time series dataset size has been increased from 100 to 200 observations, providing more stable and statistically significant results. A random seed has been set for reproducibility, and the data now better differentiates between stationary and non-stationary series.

2. The test comparing stationary versus non-stationary series now validates that the ADF statistic of the stationary series is more negative than that of the non-stationary series, rather than relying solely on p-value comparisons. Additional checks ensure that both p-values lie within the valid range of 0 to 1.

3. Several new metric identifiers have been added to the tests for individual classification metrics (e.g., AbsoluteError, BrierScore, CalibrationError, among others), expanding the coverage of unit metrics considered during testing.

4. The order of output processing has been adjusted: the MetricOutputHandler has been moved to execute last to ensure that unit metric outputs are processed after other outputs. This reordering helps maintain better control over the flow and handling of test results.

Overall, these enhancements improve the robustness and reliability of the testing framework without altering core functionalities beyond the test suite and output processing middleware.

Test Suggestions

• Run the entire test suite to ensure all new and existing tests pass with the updated dataset size and random seed.
• Explicitly verify that the ADF statistics are more negative for stationary series compared to non-stationary ones in various scenarios.
• Confirm that p-values are always between 0 and 1 for both series across multiple runs.
• Test the output processing order to ensure that MetricOutputHandler executes last and that all output handlers function as expected.
• Review the new metric identifiers to check that they integrate seamlessly with the overall metric evaluation process.