[P1.5-S5-ARI-37] Round 2 execution — tool mode + API mode

[bjridicodes]

Ticket: P1.5-S5-06

Type: Execution | Est: 2d

Goal: All 30 incidents run through ARIA on real GCP infrastructure in both invocation modes. Results captured in a structured file for AC scoring.

Scope (depends on P1.5-S5-05 test harness):

• Tool mode: direct ARIAPipeline.run(incident_number) for all 30 incidents; capture per-incident: classification_label, confidence_band, affected_ci_matched, log_recall, duration_ms
• API mode: POST /api/v1/pipeline/run for all 30 incidents via HTTP; capture: HTTP response time, notification_sent, duration_ms, AC-01 timestamp delta, AC-06 timestamp delta
• Results written to tests/acceptance/results_round2.json

Acceptance criteria:

• All 30 tool mode runs complete without pipeline crashes
• All 30 API mode runs complete without HTTP errors
• results_round2.json created with per-incident per-mode data for all 60 runs
• No unhandled exceptions in ARIA logs during execution