Issue #294

Problem

Our current peer connection implementation has a potential issue with connection cleanup when peers disconnect. When we split the TcpStream and run the read half in a detached task, we rely entirely on the peer's proper shutdown behavior to terminate the connection. If the peer doesn't close their connection cleanly (which appears to be happening, causing fin-wait2 states on Linux), the read task can remain active indefinitely, potentially leading to resource leaks like file descriptor exhaustion.

Solution

This PR implements a cancellation pattern for more reliable connection cleanup:

1. Added a oneshot channel to coordinate shutdown between the peer and its read task
2. When a peer initiates shutdown, it signals the read task through this channel
3. The read task now uses tokio::select! to respond to either:
   • Normal completion of the read loop
   • Cancellation signal from the peer

Implementation Details

• Uses tokio::sync::oneshot channel instead of CancellationToken from tokio-utils to minimize dependencies
• Follows Tokio's recommended shutdown patterns (see Tokio shutdown documentation)
• Adds graceful error handling for shutdown signal propagation

Impact

This change should prevent connection resource leaks by ensuring read tasks are properly cleaned up, even when remote peers don't follow correct TCP shutdown procedures.