Add Pareto frontier output for policy trade-off selection

[dgenio]

Context / Problem

• One score can hide useful non-dominated policy choices.
• Stakeholders need trade-off visibility before rollout.

Proposal

• Compute Pareto-optimal policies across key metrics.
• Render frontier table in report and CLI output.
• Add short legend for objective directions.

Acceptance Criteria

• Frontier table appears in markdown report.
• CLI lists frontier policy names.
• Tests validate frontier membership on fixtures.

Implementation Notes

• Use success up, unsafe down, cost down, latency down.
• Keep score ranking table unchanged.

Tasks / Checklist

• Implement frontier detection utility.
• Add report section and CLI summary lines.
• Add fixture tests for dominated and non-dominated cases.

References

• https://docs.langchain.com/langsmith/evaluation
• https://developers.openai.com/api/docs/guides/evals