[cjavad]

The multi-printer connection loop can consume a CPU core during shutdown. asyncio.CancelledError was treated as a reconnectable network error, so cancellation was swallowed. Once the event loop's DNS executor shut down, every reconnect raised RuntimeError, logged a full traceback, and retried immediately.

This change exits and cleans up on cancellation, routes unexpected connection failures through the existing reconnect backoff, and honors Retry-After: Ns from failed add_connection responses. A server response such as Retry-After: 249s now delays the next add attempt for 249 seconds instead of using an unrelated local exponential delay.

Validation:

• pytest -q tests/test_connection_retry.py tests/test_multi_printer_retry.py tests/test_connection_manager.py (7 passed)
• Remaining suite excluding live test_client_app.py: 84 passed; three environment failures require writes outside the workspace or network-interface access.
• Live strace before mitigation measured about 72 traceback cycles per second on each wedged worker.