From Silos to Systems: Process-Oriented Hazard Analysis for AI Systems