[P1.5-S5-ARI-36] Test harness design + implementation

[bjridicodes]

Ticket: P1.5-S5-05

Type: Feature | Est: TBD — scope in dedicated session

Goal: Automated harness that runs all 30 incidents in both invocation modes (tool and API), collects per-incident results, and scores AC-01 through AC-06 against ground_truth.json.

Scope: This ticket is a placeholder. Do not start implementation without a dedicated scoping session that produces an implementation spec in .docs/.

Acceptance criteria:

• Dedicated scoping session held before S5 execution starts
• Implementation spec written to .docs/test_harness_spec.md
• Harness implementation complete and tested on ≥ 3 incidents before running full 30