Backfill citation goal-start trajectory summary

[huangruiteng]

Summary

• Backfill a public-safe trajectory_public_summary block for the citation-check goal-start bridge-timeout run.
• Refresh the generated case-analysis Markdown coverage tables from compact JSON.
• Update the focused case-analysis smoke expectations from 5 to 6 trajectory summaries and assert the new goal-start summary counters.

Validation

• python3 examples/benchmark-case-analysis-smoke.py
• python3 -m json.tool docs/research/long-horizon-agent-benchmarks/benchmark-case-analysis.json
• git diff --check -- docs/research/long-horizon-agent-benchmarks/benchmark-case-analysis.json docs/research/long-horizon-agent-benchmarks/benchmark-case-analysis.md examples/benchmark-case-analysis-smoke.py
• loopx check --scan-path docs/research/long-horizon-agent-benchmarks/benchmark-case-analysis.json --scan-path docs/research/long-horizon-agent-benchmarks/benchmark-case-analysis.md --scan-path examples/benchmark-case-analysis-smoke.py

Boundary

• Used only compact/public trajectory counters from the local public trace artifact.
• Did not read or commit raw task text, raw logs, raw trajectory bodies, verifier output, credentials, uploads, or submissions.
• No benchmark jobs launched and no runner/scoring behavior changed.