feat: add non-blocking handshakes and tunnel health monitoring (#13)

[rafabd1]

Feature: Add non-blocking handshakes and tunnel health monitoring (#13)

Closes #13

Problem

The accept_loop has several resilience issues:

1. Blocking handshakes: Each incoming handshake blocks the accept loop, preventing subsequent connection attempts from being processed. If multiple peers connect simultaneously, only one handshake proceeds at a time.

2. No timeouts: Incoming handshakes can hang indefinitely if the peer becomes unresponsive during the handshake, leaving the connection in a zombie state.

3. No tunnel health monitoring: There's no way to detect when the I2P tunnel becomes degraded or unresponsive. Users receive no feedback about tunnel status.

4. No status feedback: Users have no visibility into whether the tunnel is healthy or degraded, making troubleshooting difficult.

These issues are particularly problematic on I2P, where network conditions can be unstable and tunnels often degrade temporarily.

Solution

Implement non-blocking handshakes with tunnel health monitoring:

1. Non-blocking handshakes:

   • Spawn each handshake in a separate tokio task
   • Accept loop continues to process new connections while handshakes run concurrently
   • Prevents multiple connection attempts from queuing indefinitely

2. Handshake timeouts:

   • 60s timeout for all incoming handshakes
   • Emits connection_error event on timeout
   • Prevents zombie connections from hanging indefinitely

3. Tunnel health monitoring:

   • Background tunnel_health_monitor task runs in parallel with accept_loop
   • Checks SAM connectivity every 30 seconds
   • Quick health check via SAM HELLO command (non-invasive)

4. Health status events:

   • tunnel_healthy: Emitted when tunnel is responsive
   • tunnel_degraded: Emitted when tunnel is unresponsive
   • Frontend can display status indicators to users

Implementation details

• Health monitor is spawned in a separate task when accept_loop starts
• Both tasks exit gracefully when the session is replaced or dropped
• Health checks use simple SAM HELLO to verify connectivity
• Timeout uses tokio::time::timeout for clean cancellation
• All operations are non-blocking and asynchronous

Benefits

• Accept loop remains responsive even during slow handshakes
• Multiple connection attempts can be processed concurrently
• Users receive feedback about tunnel health status
• Degraded tunnels are detected quickly (within 30s)
• No zombie connections from hanging handshakes