feat(tooling): benchmark summary report across models/metrics (external-first)

[christso]

Canonical Plan

This issue body is the source of truth for implementation.

Objective

We generate consolidated benchmark summaries across models and metrics from existing result artifacts.

Why This Matters

• Benchmark interpretation currently requires stitching multiple outputs manually.
• A single consolidated report lowers effort and improves repeatability.
• Teams can communicate results faster across engineering and product stakeholders.

Why This Location

• This is a synthesis/reporting concern over existing outputs.
• We can deliver value without expanding runtime/evaluator surface area.
• External-first keeps reporting flexible and decoupled from execution.

Architecture Boundary

External-first reporting/tooling.

Deliverable Location

• Primary location (in-repo tooling path): examples/features/benchmark-tooling/
• Script location: examples/features/benchmark-tooling/scripts/ (benchmark report utility)
• Usage/docs: examples/features/benchmark-tooling/README.md

Design Latitude

We choose command shape, report format defaults, and aggregation presentation based on usability and maintainability.

Acceptance Signals

• We can produce one benchmark-oriented summary from multiple result files.
• We include per-target/per-metric aggregates and uncertainty when available.
• We support both machine-readable and human-readable outputs.

Non-Goals

• No evaluator runtime changes.
• No requirement to add a new core command if a tooling path is cleaner.