[Story]: Region-pinned inference routing enforcement

## Summary
Implement region-pinned routing so an inference request tagged with a residency policy (e.g. EU-only) is routed only to infrastructure or upstream endpoints satisfying that policy.

## Intent / Source of truth
This is the enforcement mechanism the epic's documented boundary claim depends on. Part of [Epic] Data residency selection and enforcement.

## Scope
- [ ] Residency-policy tag on requests (per org/BU, wired through the org-hierarchy policy model where available)
- [ ] Routing logic in the gateway restricting requests to region-compliant backends
- [ ] Rejection (not silent fallback) when no region-compliant backend exists for a requested model

## Out of scope
- Building new regional backend infrastructure (this story routes to what exists/is offered upstream)

## Verification
Test asserting an EU-pinned request is never routed to a non-EU backend, and is explicitly rejected (with a clear error) rather than silently falling back, when no compliant backend exists.

## Risk assessment
A silent fallback to a non-compliant region would be a serious compliance failure disguised as a working feature — the rejection path must be the default, not an edge case.

## AI Usage Declaration
Drafted with AI assistance during the 2026-08-13 cross-repo backlog consolidation and enterprise-readiness research. A human owns intent, verification and consequences.