Tiered rate limiter with limits that change at runtime

Build a rate-limiting service in Python. Each client belongs to a tier (for example free, pro, enterprise), and each tier has its own limit. A request arrives with a client id, and the service decides whether to allow it under that client's tier.

1. Implement the limiter and its interface.
2. Let an operator change a tier's limits while the service runs, without a restart, and define what happens to clients mid-window.
3. Write tests for it.
4. Discuss how you would scale it beyond one process: shared state, accuracy versus latency, and failure behavior.