Return aggregated internal quality metrics (last 7 days).

Reads from .sdd/metrics/ JSONL files to compute:

• per_model: per-model success rate, avg tokens, and completion time distribution (p50/p90/p99).
• overall: aggregate across all models.
• gate_stats: per-gate pass/blocked/flagged counts (last 30 days).
• guardrail_pass_rate: fraction of gate checks that passed.
• review_rejection_rate: fraction of tasks that failed overall.

Returns an empty structure when no metric data exists yet.