Add long-horizon Arazzo goal evaluation

[SonAIengine]

Summary

• add a public outcome evaluator and replayable long-horizon baseline
• benchmark paired 1,000-tool catalogs with and without Arazzo across 3, 10, and 30-step workflows
• record target, required-tool, ordering, binding, goal-state, latency, and token-budget metrics
• document the methodology and expose focused Make targets

Results

• Arazzo: target/required-tool/order/binding/goal metrics all 1.00
• OpenAPI-only control: order/binding/goal metrics 0.00
• average admitted schema tokens: 260.67; maximum: 301

This benchmark is deterministic (model=none) and does not claim LLM planning quality.

Validation

• poetry run ruff check .
• poetry run ruff format --check .
• poetry run pytest tests/ -q (1235 passed, 47 skipped)
• frozen Arazzo long-horizon benchmark result passes its gates