pipeline destroy do not rely on stop propagate signal

[GaijinKa]

Description

This PR refines the lifecycle management of nodes and pipelines within the Juturna core library. The primary focus is on implementing a robust shutdown mechanism that ensures data integrity by "draining" nodes before they fully stop. By introducing a join() method and using threading conditions to track pending updates, the system now waits for active tasks to complete before releasing resources.

These changes address several race conditions and resource management issues during pipeline termination, specifically targeting the improvements discussed in issue #160.

PR type

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Code refactoring
• Performance improvement
• Test update
• Build/CI configuration change
• Other (please describe):

---

Key modifications and changes

• Enhanced Thread Safety & Synchronization:
  • Added a _pending_condition (threading.Condition) and a _pending_updates counter to the Node class to track in-flight data processing.
  • Implemented a join() method in Node to safely wait for internal threads (source, worker, update) to terminate, ensuring they are not joined by the current thread.
  • Fixed a bug in the synchroniser setter where it incorrectly assigned to the public property name instead of the private _synchroniser attribute.
• Improved Shutdown Logic:
  • Introduced a _draining event to signal when a node is stopping, allowing it to discard new incoming messages while finishing current work.
  • Modified Pipeline.stop() to iterate through nodes layer-by-layer (BFS) and explicitly call join() on each node after sending a STOP signal.
  • Updated Buffer.get and internal node loops to use timeouts (0.1s) instead of blocking indefinitely, allowing threads to check for stop events more frequently.
• Message & Signal Handling:
  • Fix a bug where node communication for SUSPEND and RESUME signals were not wrapped in a Message and ControlPayload objects.
• State & Resource Management:
  • Added DESTROYED to PipelineStatus and ensured the pipeline status is updated upon calling the destroy() method.
  • Simplified PipelineManager.stop_pipeline by removing redundant manual status updates, delegating state management to the pipeline itself.

---

Affected components

• Juturna Core Library:
  • juturna.components (_buffer.py, _node.py, _pipeline.py, _pipeline_manager.py)

---

Additional context

The transition from blocking get() calls to timeout-based loops significantly reduces the risk of "zombie" threads remaining active after a pipeline stop command. By tracking _pending_updates with a Condition variable, the node now ensures that any call to the update() method is fully completed before the thread is joined and resources are released.