Link traced evaluation runs to their dataset examples

Problem

Successful traced evaluation runs set referenceExampleId only when EvaluateErrorHandling.LOG is configured. When using IGNORE, successful runs are not linked to the dataset example that produced them.

This weakens traceability and makes it harder to inspect evaluation results by example in LangSmith.

Expected behavior

All successful traced evaluation runs should include the ID of their source dataset example, regardless of error-handling mode.

Proposed fix

Always set referenceExampleId to example.id() when creating the evaluation target’s TraceConfig.

Acceptance criteria

• Traced runs include reference_example_id for LOG and IGNORE modes.
• Existing error-handling behavior is unchanged.
• Add regression coverage for a successful evaluation using EvaluateErrorHandling.IGNORE.

Start at the code that creates the evaluation target’s TraceConfig and compare how referenceExampleId is set for EvaluateErrorHandling.LOG versus IGNORE. Use example.id() as the expected source identifier, then add regression coverage for a successful IGNORE evaluation. Done means traced runs include reference_example_id in both modes while existing error-handling behavior remains unchanged.