netstack: revert SND.NXT when purging the write queue

There are only 3 places we remove from the write queue. The other two are
fairly self-contained and don't look suspicious. They are:

- tcp.sender.maybeSendSegment - removes iff the segments are merged
- tcp.sender.handleRcvdSegment - removes iff the whole segment is covered by an
  incoming ACK

Given that the panic occurs when the write queue is empty and SND.NXT !=
SND.UNA, the bug likely occurs when either the writeList removes a segment or
SND.NXT increments.

PiperOrigin-RevId: 698885878

pkg/tcpip/transport/tcp/endpoint.go

@@ -1013,16 +1013,13 @@ func (e *Endpoint) purgeWriteQueue() {
 		e.sndQueueInfo.sndQueueMu.Lock()
 		defer e.sndQueueInfo.sndQueueMu.Unlock()
 		e.snd.updateWriteNext(nil)
-		for {
-			s := e.snd.writeList.Front()
-			if s == nil {
-				break
-			}
+		for s := e.snd.writeList.Front(); s != nil; s = e.snd.writeList.Front() {
 			e.snd.writeList.Remove(s)
 			s.DecRef()
 		}
 		e.sndQueueInfo.SndBufUsed = 0
 		e.sndQueueInfo.SndClosed = true
+		e.snd.SndNxt = e.snd.SndUna
 	}
 }