Logged target errors abort dataset evaluation

Problem

EvaluateErrorHandling.LOG is intended to log target failures and include errored runs in the experiment. However, EvaluateRunner captures the active RunTree only after the target returns successfully.

When a target throws, runTree remains null. Although tracing records and rethrows the exception, the evaluation runner then throws IllegalStateException rather than retaining the errored run.

Expected behavior

With EvaluateErrorHandling.LOG, a target exception should create an errored run associated with its dataset example and allow the evaluation to continue.

Proposed fix

Capture the active RunTree before invoking the target function.

Acceptance criteria

• A throwing target produces an evaluation row with LOG handling.
• The row’s run includes the target exception.
• The run remains linked to its source example.

Start in EvaluateRunner and trace how the active RunTree is captured around target invocation and how EvaluateErrorHandling.LOG processes failures. Add or update evaluation coverage for a throwing target, then verify that the evaluation continues with a row whose run contains the exception and remains linked to its dataset example.