Function record_rate_limit 

pub async fn record_rate_limit(
    agent_name: &str,
    model: Option<&str>,
    error_message: &str,
)

Record a rate-limit event for an agent+model.

If the error message advertises a relative usage window (e.g. “5-hour usage limit”), the cooldown is applied to the specific model for that window length. The quota is model-scoped, so the whole agent is not cooled. If no window is detected, fall back to the generic agent+model exponential backoff path.