exec: diagnose connection failures with server health checks

[anish749]

Problem

When the WebSocket connection to the Jupyter server drops (VPN disconnect, network interruption, proxy timeout, kernel shutdown), the CLI shows a generic "Connection is already closed." error from the websocket library. The caller has no way to tell what actually happened.

Solution

Added a diagnostics module (session/diagnostics.py) that:

1. Classifies exceptions as connection-related vs code execution errors
2. For connection errors, performs a follow-up HTTP health check against the Jupyter server (with a 5s timeout) to determine the cause
3. Returns a specific, actionable error message:
   • Server unreachable → "Check network connectivity and VPN"
   • Kernel gone (404) → "Kernel is no longer running... Create a new session"
   • Kernel still alive → "WebSocket connection lost but kernel is still running... likely a network interruption"
   • Server returning errors → reports the HTTP status code

The Session.execute() exception handler now uses these diagnostics for connection failures, while non-connection errors (e.g. kernel code exceptions) continue to pass through unchanged.