Add additional buckets to fanout histogram (grafana#685)

- The prometheus_fanout_latency is used to measure latencies for data
    moving between components.
  - The default buckets are limited to 10s, which limits the ability
    detect very high latencies.
  - Adds to additional buckets (30s, 60s) for additional resolution.
  - Note: I initially planned to replace some of the existing buckets
    instead of adding new ones, but I believe most components will fall
    into the smaller buckets and reducing the resolution there would
    lead to distortions in the quantile calculations (and potentially
    create the ilusion of performance degradation). Note however that
    the 2 new buckets will result in increased cardinality of 2 metrics
    per collector instance.

internal/component/prometheus/fanout.go

@@ -33,8 +33,9 @@ type Fanout struct {
 // NewFanout creates a fanout appendable.
 func NewFanout(children []storage.Appendable, componentID string, register prometheus.Registerer, ls labelstore.LabelStore) *Fanout {
 	wl := prometheus.NewHistogram(prometheus.HistogramOpts{
-		Name: "prometheus_fanout_latency",
-		Help: "Write latency for sending to direct and indirect components",
+		Name:    "prometheus_fanout_latency",
+		Help:    "Write latency for sending to direct and indirect components",
+		Buckets: []float64{.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10, 30, 60},
 	})
 	_ = register.Register(wl)