Move processing heartbeats off the reactor

[crmne]

Closes #7.

This moves the processing heartbeat / abandoned-job recovery loop off the job reactor and onto a dedicated Ruby thread with its own Redis client.

The important behavior change is that a worker can now continue to own an in-flight job even if the job body temporarily blocks the reactor. Throughput may degrade under blocking workloads, but correctness should not: a still-running job should not look abandoned and be re-executed.

What changed:

• ProcessingList#start now runs the heartbeat/requeue loop in a dedicated thread.
• The thread uses its own Async::Redis::Client, so the liveness path no longer depends on the job reactor yielding.
• Server exposes configurable heartbeat_interval and heartbeat_factor options and shuts the heartbeat thread down cleanly.
• Tests were updated to cover the threaded heartbeat path, and the existing status-string assertion was made less timing-sensitive.

Validation:

• bundle exec sus