Token bucket depletion starves queues during burst traffic

Token bucket throttling depletes when burst size exceeds bucket_size capacity, starving queues. Occurs in 15% of IoT spike scenarios. Root cause: fixed fill_rate cannot adapt to load variance, causing legitimate tasks to be throttled during bursts.

Implement adaptive throttling using feedback loop: query Prometheus for queue length (threshold: 1k messages), then dynamically adjust fill_rate (e.g., halve rate when threshold exceeded). Tune bucket capacity to 2x expected peak burst size.