Use persistent read-only manual monitoring

[temiroff]

What changed

• Replace repeated native one-shot reads with one managed read-only joint stream.
• Add transport-aware cleanup and stale-session replacement.
• Add a live_monitor input so deployed leader release can avoid a redundant monitor.
• Bump package version to 0.3.1 and require blacknode-ros2 0.5.6.

Why

The previous 10 Hz create/destroy loop left stale ROS discovery entries and could look like competing processes.

Validation

• pytest -q packages/blacknode-controllers/tests (60 passed)