Unlock tcp endpoint mutex before blocking forever.

This was occasionally causing tests to get stuck due to races with the save
process, during which the same mutex is acquired.

PiperOrigin-RevId: 340789616

pkg/tcpip/transport/tcp/connect.go

@@ -1370,7 +1370,9 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{
 	drained := e.drainDone != nil
 	if drained {
 		close(e.drainDone)
+		e.mu.Unlock()
 		<-e.undrain
+		e.mu.Lock()
 	}
 
 	// Set up the functions that will be called when the main protocol loop