Online Evaluation

[mmabrouk]

We would like to add the ability to automatically evaluate their AI applications in production to:

• Monitor model performance in real-time
• Detect quality degradation early
• Make data-driven decisions about model improvements
• Ensure consistent quality across production traffic

We would add the possibility for the user to:

• Set up a trace filter to run on new and historical data
• Configure an evaluator (llm-as-a-judge, code, webhook, or a built-in evaluator)

And that evaluator will be run automatically on the traces

[mmabrouk]

Issues: #2722
#2723
#2724