mapper: preserve node pings when a full update is queued

Updates #3523

(cherry picked from commit 17cb110892)
(cherry picked from commit 7c2260c602)
(cherry picked from commit 6875c4bc8, test only)

hscontrol/mapper/node_conn.go

@@ -4,6 +4,7 @@ import (
 	"errors"
 	"fmt"
 	"reflect"
+	"slices"
 	"strconv"
 	"sync"
 	"sync/atomic"
@@ -291,6 +292,14 @@ func (mc *multiChannelNodeConn) prependPending(changes ...change.Change) {
 	mc.pendingMu.Unlock()
 }
 
+// collapsePendingToFull replaces pending, together with incoming, by a
+// single full update and the pings it cannot carry ([change.CollapseToFull]).
+func (mc *multiChannelNodeConn) collapsePendingToFull(incoming []change.Change) {
+	mc.pendingMu.Lock()
+	mc.pending = change.CollapseToFull(mc.id, slices.Concat(mc.pending, incoming))
+	mc.pendingMu.Unlock()
+}
+
 // drainPending atomically removes and returns all pending changes.
 // Returns nil if there are no pending changes.
 func (mc *multiChannelNodeConn) drainPending() []change.Change {