fix: tunnel dying some time after a network handover

[vemneyy]

Introduced in my last commit tunFdForCore() hands the core vpnInterface.dup() on the premise that the core closes the descriptor it was given and would otherwise take the VPN interface down. That premise does not hold. In the pinned xray-core, AndroidTun.Close() (proxy/tun/tun_android.go) returns nil without touching the fd and the gVisor fdbased endpoint it is passed to has an empty Close() reading stops through Attach(nil) → dispatcher.Stop() → Wait(). Nothing closes it so the duplicate just becomes a ParcelFileDescriptor nobody keeps a reference to and its finalizer performs the only close that fd ever gets at whatever point the GC runs, on a descriptor the running core is still dispatching on. From then on nothing arrives from the tun and every app hangs while the core stays up and only restarting the service recovers. Because the timing is the GC's the failure is not aligned with the handover that caused it which is what makes it hard to trace

[eliotcougar]

GPT5.6-Sol-xHigh fully agrees with the analysis and the solution.