Use histogram, not summary, in WAI

https://prometheus.io/docs/practices/histograms/ has a full discussion,
but the gist is that histograms can be aggregated across processes whereas
summaries cannot.

This makes histograms more useful in environments where there is more than one
process handling web requests, which is perhaps the more common case.

wai-middleware-prometheus/src/Network/Wai/Middleware/Prometheus.hs

@@ -49,9 +49,9 @@ instance Default.Default PrometheusSettings where
 {-# NOINLINE requestLatency #-}
 -- XXX: https://prometheus.io/docs/practices/naming/ says this should be
 -- _seconds, not _microseconds.
-requestLatency :: Prom.Metric (Prom.Vector Prom.Label3 Prom.Summary)
+requestLatency :: Prom.Metric (Prom.Vector Prom.Label3 Prom.Histogram)
 requestLatency = Prom.unsafeRegisterIO $ Prom.vector ("handler", "method", "status_code")
-                                       $ Prom.summary info Prom.defaultQuantiles
+                                       $ Prom.histogram info Prom.defaultBuckets
     where info = Prom.Info "http_request_duration_seconds"
                            "The HTTP request latencies in seconds."