Support parallel execution for evaluation command

[Leward]

Running evaluations in the current neuron-core/neuron-ai repository is slow because the evaluation command processes evaluators and dataset items strictly sequentially.

Details

• The EvaluationCommand runs each evaluator one by one.
• Within each evaluator, dataset items are processed in sequence via EvaluatorRunner::run().
• For larger datasets or many evaluators, this becomes a bottleneck.

[ilvalerione]

Just to update about the latest improvements to the Evaluation module in the latest release 3.2.5. It now supports output interfaces. It's a more extensible architecture to make contributions easier, or let developers free to implement their custom output components. Feel free to provide any feedback if you want, we are continuing to work on this.