Add baseline comparison with regression detection

[aviralgarg05]

Summary

Implement baseline comparison that fails CI if metrics regress beyond a threshold.

Tasks

• Add agentunit baseline save to store current results as baseline.
• Add agentunit eval --fail-on-regression to compare against baseline.
• Support configurable thresholds per metric.
• Output diff report showing improvements/regressions.

Acceptance criteria

• Baseline file stored in .agentunit/baseline.json.
• CI fails if any metric drops more than threshold.
• Clear diff output showing metric changes.