session-extend: drop dead retry loop in mgmClient.ExtendAuthSession

The backoff loop only retried on codes.Canceled, but mgmCtx was derived
from context.Background() and never cancelled by anything — so every
real error path (Unavailable, DeadlineExceeded, etc.) went through
backoff.Permanent on the first attempt. The loop was a no-op wrapper
that just held the call open for the daemon's lifetime regardless of
shutdown.

Replace with a single context.WithTimeout(c.ctx, ConnectTimeout) call.
Daemon shutdown now interrupts the RPC; behaviour on real errors is
unchanged.

shared/management/client/grpc.go

@@ -630,26 +630,14 @@ func (c *GrpcClient) ExtendAuthSession(sysInfo *system.Info, jwtToken string) (*
 		return nil, err
 	}
 
-	var resp *proto.EncryptedMessage
+	mgmCtx, cancel := context.WithTimeout(c.ctx, ConnectTimeout)
-	operation := func() error {
+	defer cancel()
-		mgmCtx, cancel := context.WithTimeout(context.Background(), ConnectTimeout)
-		defer cancel()
 
-		var err error
+	resp, err := c.realClient.ExtendAuthSession(mgmCtx, &proto.EncryptedMessage{
-		resp, err = c.realClient.ExtendAuthSession(mgmCtx, &proto.EncryptedMessage{
+		WgPubKey: c.key.PublicKey().String(),
-			WgPubKey: c.key.PublicKey().String(),
+		Body:     reqBody,
-			Body:     reqBody,
+	})
-		})
+	if err != nil {
-		if err != nil {
-			if s, ok := gstatus.FromError(err); ok && s.Code() == codes.Canceled {
-				return err
-			}
-			return backoff.Permanent(err)
-		}
-		return nil
-	}
-
-	if err := backoff.Retry(operation, nbgrpc.Backoff(c.ctx)); err != nil {
 		log.Errorf("failed to extend auth session on Management Service: %v", err)
 		return nil, err
 	}