Route AI workloads by data residency, sensitivity, user region, retention policy, and allowed processors