Write an extremely long, exhaustive, deeply detail...

Write an extremely long, exhaustive, deeply detailed technical guide on building a production-grade distributed task queue from scratch in Python. Do not stop early, do not summarize, do not abbreviate. Cover: architecture overview, broker design, serialization formats, at-least-once vs exactly-once semantics, idempotency keys, visibility timeouts, retries with exponential backoff and jitter, dead letter queues, priority queues, fair scheduling across tenants, worker pool models (threads, processes, asyncio), graceful shutdown, heartbeats and liveness detection, horizontal scaling, backpressure, rate limiting, observability (metrics, tracing, structured logging), storage schema with full SQL DDL, complete runnable code for the broker, worker, scheduler, and client, a full test suite, Docker and Kubernetes manifests, benchmarking methodology, failure mode analysis, and a migration plan from Celery. Include complete code for every component with no placeholders and no truncation. Write as much as you possibly can.