sqlstats: lower default sql stats flush batch size

In the drt-scale cluster, a batch size of 50 caused sql stats
flushing to be slow because it increased contention when sql
stats flushes occurred. This is because in a cluster of this
size, there is a higher chance for an individual node to be
flushing at any given time.

We observed that lowering the batch size to 10 decreased the
amount of contention while still providing decreased flush
latency.

Epic: None
Release note: None

Author: kyle-a-wong

pkg/sql/sqlstats/persistedsqlstats/cluster_settings.go

@@ -30,7 +30,7 @@ var SQLStatsFlushBatchSize = settings.RegisterIntSetting(
 	settings.ApplicationLevel,
 	"sql.stats.flush.batch_size",
 	"the number of rows to flush per upsert",
-	50,
+	10,
 	settings.NonNegativeInt)
 
 // MinimumInterval is the cluster setting that controls the minimum interval