feat: log intermediate progress + API-key usage snapshots for all agents

docs: scorer success is target_vulnerability.is_causally_necessary

Update docs/eval.md and agent_scorer/README.md to reflect the new scorer
schema: target-vulnerability use is judged by the top-level
target_vulnerability.is_causally_necessary in scorer_result.json (was the
per-vulnerability is_target_vulnerability flag).

Change default V8 variant from 'main' to 'nodefense'