Add eval-harness cost attribution estimator

[orban]

Summary

• carry Claude-reported USD cost through task and AGENTbench execution results
• add component-level cost attribution to eval-harness JSON/Markdown reports and resume summaries
• add focused tests and README docs for fix vs skill_generation cost estimates

Testing

• pytest eval-harness/tests/test_reporter.py eval-harness/tests/test_task_runner.py eval-harness/tests/test_resume.py eval-harness/tests/test_agentbench.py -q
• pytest eval-harness/tests -q (fails only in docker-dependent test_docker_runner because the local Docker socket is unavailable in this environment)

[orban]

Superseded by #31, which replays this implementation on top of the edge-case and resumed-run median fixes. Closing to consolidate review.