fix(sync): prevent concurrent tail movements (#313)

Moving tail(or pruning) may take a lot of time, while it happens, new
network heads can arrive, interfering with the in-progress move. This
change simply detects that and skips tail processing if there is one in
progress

Author: Wondertan

sync/syncer.go

@@ -47,6 +47,8 @@ type Syncer[H header.Header[H]] struct {
 	pending ranges[H]
 	// incomingMu ensures only one incoming network head candidate is processed at the time
 	incomingMu sync.Mutex
+	// tailMu prevents concurrent tail movements
+	tailMu sync.Mutex
 
 	// controls lifecycle for syncLoop
 	ctx    context.Context

sync/syncer_tail.go

@@ -18,6 +18,16 @@ func (s *Syncer[H]) subjectiveTail(ctx context.Context, head H) (H, error) {
 		return oldTail, err
 	}
 
+	if !s.tailMu.TryLock() {
+		// prevents concurrent tail estimation and moving
+		//
+		// If a new head arrives, while tail for the previous head is still in progress
+		// it is valid to skip tail renewal for the new one. It will be resolved with a more recent head
+		// once in progress tail finishes.
+		return oldTail, nil
+	}
+	defer s.tailMu.Unlock()
+
 	newTail, err := s.renewTail(ctx, oldTail, head)
 	if err != nil {
 		return oldTail, fmt.Errorf("updating tail: %w", err)