Fix timeout evaluations getting positive fitness

[bud333]

Summary

Fix timeout evaluation paths so timed-out programs cannot receive positive fitness.

In the current behavior, timeout results may omit combined_score, and the fallback fitness logic can incorrectly treat boolean flags like timeout=True as numeric values. This allows failed or timed-out programs to receive a positive fitness score.

For example, a timeout result like {"error": 0.0, "timeout": True} could previously receive a positive fallback fitness because True was treated as 1.0. This change makes timeout paths return combined_score=0.0 explicitly and excludes boolean flags from fallback fitness aggregation.

Changes

• return combined_score=0.0 for direct evaluation timeouts
• return combined_score=0.0 for cascade stage timeouts
• set a meaningful string error message for timeout cases
• exclude boolean values from numeric aggregation in fitness fallback logic
• add regression tests for timeout scoring and bool exclusion

Why

This fixes a case where timed-out programs could receive positive fitness when timeout results omitted combined_score and boolean flags were treated as numeric values in fallback
fitness calculation.

Verification

Passed:

• export OPENAI_API_KEY=test-key-for-unit-tests && python -m unittest tests.test_evaluator_timeout tests.test_metrics_utils
• export OPENAI_API_KEY=test-key-for-unit-tests && python -m unittest discover tests

[CLAassistant]

All committers have signed the CLA.