MCP server exits cleanly when its parent disconnects — no more zombie processes

What shipped

• •Added handlers for SIGTERM, SIGINT, SIGHUP, stdin 'end', stdin 'close', and the MCP transport's onclose hook. Any of these now triggers a graceful exit.
• •Replaced the previous uncaughtException handler that swallowed fatal errors and kept running. The handler now logs and exits with code 1, so a crashed server actually dies and Claude Desktop respawns it cleanly on the next call. Same treatment for unhandledRejection.
• •Added a parent-alive watchdog that polls process.ppid every 30 seconds and exits if the parent has been reaped (process.ppid === 1 means we're orphaned to launchd). Belt-and-braces for the rare case where stdio events don't fire on abnormal parent termination. Uses .unref() so it can never keep the process alive on its own.
• •Shutdown is idempotent and bounded — a 200ms grace window after triggering, then process.exit. If anything is keeping the event loop alive (timers, sockets, pending I/O), the hard exit ensures we still terminate.
• •Verified locally: SIGTERM now produces '[Orbit] shutdown: SIGTERM' on stderr and exits with code 0 within ~200ms, vs the previous behaviour of running forever.