feat(copilot): stabilize concurrency and enhance terminal context management

- AI Concurrency:
  - Implemented a dedicated background event loop (ConnpyAILoop) in a separate thread for AI tasks to ensure thread safety and event loop affinity.
  - Added 'run_ai_async' utility to funnel all LiteLLM calls through the dedicated loop.
  - Implemented global 'cleanup()' for safe closure of sync/async LiteLLM sessions.

- gRPC & Remote Sessions:
  - Enhanced 'NodeServicer' to identify command blocks within the terminal buffer using prompt regex/byte tracking.
  - Added support for selective context retrieval via 'context_start_pos' in the gRPC Interact stream.
  - Synchronized remote Copilot behavior by enriching questions with session history (last 5 queries) in 'NodeStub'.
  - Optimized token usage by cleaning 'node_info' metadata before AI transmission.

- Terminal Context & Core:
  - Modified 'node.connect' to always initialize 'mylog' (BytesIO) buffer regardless of disk logging configuration, ensuring Copilot context availability.
  - Integrated 'ai.cleanup()' in CLI (connapp) and Server (api) exit points for graceful shutdowns.
  - Suppressed LiteLLM internal streaming coroutine warnings during task cancellation.

connpy/connapp.py

@@ -470,6 +470,13 @@ class connapp:
             # Handle global Ctrl+C gracefully
             printer.warning("Operation cancelled by user.")
             sys.exit(130)
+        finally:
+            # Safely cleanup AI sessions (litellm)
+            try:
+                from .ai import cleanup
+                cleanup()
+            except ImportError:
+                pass
 
     class _store_type(argparse.Action):
         #Custom store type for cli app.