fix: TRAC-1072 Seed idle queue gauges at zero instead of deleting

TRAC-1066 fixed cloudflare_queue_backlog_messages and related gauges
freezing at a stale non-zero value by calling DeletePartialMatch on
all 7 queue gauges before repopulating them each scrape. This
overcorrected: since Cloudflare's queueBacklogAdaptiveGroups and
queueConsumerMetricsAdaptiveGroups only emit rows for queues with
activity in the scrape window, an idle queue's series got wiped and
never repopulated, leaving Grafana with no data for hours at a time
instead of a correct zero baseline.

For the 3 gauges keyed only by queue_name/account (backlog messages,
backlog bytes, consumer concurrency), seed every currently known queue
at zero (from fetchQueueNames) immediately after the DeletePartialMatch
calls and before applying this scrape's real values. Idle queues now
read a continuous zero instead of disappearing. The 4 operation-metric
gauges keep the delete-only behavior since their action_type/outcome
label combinations aren't enumerable ahead of time.

Fixes TRAC-1072
Co-Authored-By: Claude <noreply@anthropic.com>

Author: chanceaclark

prometheus.go

@@ -815,10 +815,8 @@ func fetchQueueAnalytics(account cfaccounts.Account, wg *sync.WaitGroup, deniedM
 
 	accountName := strings.ToLower(strings.ReplaceAll(account.Name, " ", "-"))
 
-	// Drop this account's previous label combinations so queues that have gone
-	// idle (or no longer exist) disappear instead of holding their last-seen
-	// value forever — Cloudflare's queueBacklogAdaptiveGroups only emits rows
-	// for queues with activity in the scrape window.
+	// Drop this account's previous label combinations so queues that no longer
+	// exist disappear instead of holding their last-seen value forever.
 	queueBacklogMessages.DeletePartialMatch(prometheus.Labels{"account": accountName})
 	queueBacklogBytes.DeletePartialMatch(prometheus.Labels{"account": accountName})
 	queueConsumerConcurrency.DeletePartialMatch(prometheus.Labels{"account": accountName})
@@ -827,6 +825,25 @@ func fetchQueueAnalytics(account cfaccounts.Account, wg *sync.WaitGroup, deniedM
 	queueOperationLagTime.DeletePartialMatch(prometheus.Labels{"account": accountName})
 	queueOperationRetryCount.DeletePartialMatch(prometheus.Labels{"account": accountName})
 
+	// Seed every currently known queue at zero before applying this scrape's
+	// real values. Cloudflare's queueBacklogAdaptiveGroups/
+	// queueConsumerMetricsAdaptiveGroups only emit rows for queues with
+	// activity in the scrape window, so without this an idle queue would
+	// disappear from /metrics entirely instead of reading zero. The 4
+	// operation-metric gauges below aren't seeded because their action_type/
+	// outcome label combinations aren't known ahead of time.
+	for _, qName := range names {
+		if !deniedMetricsSet.Has(queueBacklogMessagesMetricName) {
+			queueBacklogMessages.With(prometheus.Labels{"queue_name": qName, "account": accountName}).Set(0)
+		}
+		if !deniedMetricsSet.Has(queueBacklogBytesMetricName) {
+			queueBacklogBytes.With(prometheus.Labels{"queue_name": qName, "account": accountName}).Set(0)
+		}
+		if !deniedMetricsSet.Has(queueConsumerConcurrencyMetricName) {
+			queueConsumerConcurrency.With(prometheus.Labels{"queue_name": qName, "account": accountName}).Set(0)
+		}
+	}
+
 	resolveQueueName := func(queueID string) string {
 		if name, ok := names[queueID]; ok {
 			return name