per-tenant rate limiting and token quota enforcement

[yai-dev]

Summary

Add first-class per-tenant rate limiting and token quota enforcement for hosted Agentrail deployments.

Current State

The current host/app layer does not yet enforce tenant-level request budgets or token quotas.

Proposed Design

1. Define a rate-limiter contract in the current host/app surface
2. Check tenant quotas before LLM execution and reconcile actual usage after completion
3. Support configurable enforcement policies such as reject, queue, or degrade
4. Resolve tenant identity from explicit run context and route headers

Affected Areas

• host/app request handling in packages/app
• runtime usage accounting in packages/core
• optional external persistence adapters as follow-up work

Acceptance Criteria

• Rate-limiter contract is defined against the current app/host architecture
• In-memory implementation exists for local/single-instance deployments
• Enforcement modes are implemented and tested
• Tenant identity resolution is documented and tested
• Metrics/events are emitted for rejection, queueing, and budget state