Rate limits

Per-tenant rate limiting to protect the AI budget. Configure requests per minute (RPM) and tokens per minute (TPM) limits.