Add worker lifecycle controls

[cmilesio]

What

Adds PauseWorkers and ResumeWorkers to the high-level queue runtime so applications can stop consumer intake without closing producer handles.

• In-memory worker pools finish active handlers, hold accepted jobs behind a lifecycle gate, and drain those jobs during graceful shutdown.
• External drivers stop and recreate only the consumer side, restoring handlers, worker counts, observers, and context decoration on resume.
• SQLite, MySQL, and Postgres now use separate producer and worker runtimes, so durable rows remain pending during both startup and live pauses.
• Core NATS returns timer-owned delayed work to the broker when a consumer stops instead of making maintenance wait for the full delay.
• Timed-out pause attempts remain reconcilable: a retry can finish the pause, and resume cannot report success while a backend is still stopped or gated.

Why

Operational maintenance needs to stop work before another backend claim without turning leases into failures, losing accepted in-memory jobs, or making dispatch unavailable. A worker-level lifecycle gives applications one reliable contract across local, SQL, and external queue drivers while preserving normal dispatch behavior and existing queue handles.

[cmilesio]

Maintenance mode is now intentionally scoped to HTTP traffic, so GoForj no longer depends on these queue lifecycle changes. Closing this PR rather than landing an unrelated queue redesign; any future worker controls should be designed and reviewed independently.