/** * Cross-session rate-limit guard for the CLEO LLM credential layer. * * Writes rate-limit state to a shared file so ALL CLEO processes * (sentient daemon, CLI calls, auxiliary router) see the same cooldown — * preventing pile-on retries from independent sessions. * * Port of Hermes' `agent/nous_rate_guard.py` (generalized for any provider). * Each 429 from a provider can trigger multiple SDK retries multiplied by * multiple Hermes retries. By recording state on the first 429 and checking * before subsequent attempts, we eliminate retry amplification across sessions. * * State is keyed on `(provider, label)` and stored at: * `${getCleoHome()}/rate-limit-state/-