Improve average block time query

This change limits the results that Postgres uses to evaluate the
window. Only `WHERE`, `HAVING`, and `GROUP BY` are used to constrain an
`OVER` (not `LIMIT`), so the change to the subquery in the `FROM` clause
effectively limits the result set to the number of blocks to include in
the average result plus one.

We could also use the (timestamp of the most recent + the timestamp of
the oldest block) / the count of records returned, but that query
requires casts from interval to integer and back to achieve the same
result. The window query ends up being similar in how it communicates.

apps/explorer/lib/explorer/chain/statistics.ex

@@ -13,10 +13,9 @@ defmodule Explorer.Chain.Statistics do
   @average_time_query """
     SELECT coalesce(avg(difference), interval '0 seconds')
     FROM (
-      SELECT timestamp - lag(timestamp) over (order by timestamp) as difference
-      FROM blocks
-      ORDER BY number DESC
-      LIMIT 100
+      SELECT b.timestamp - lag(b.timestamp) over (order by b.timestamp) as difference
+      FROM (SELECT * FROM blocks ORDER BY number DESC LIMIT 101) b
+      LIMIT 100 OFFSET 1
     ) t
   """
 

apps/explorer/test/explorer/chain/statistics_test.exs

@@ -27,11 +27,14 @@ defmodule Explorer.Chain.StatisticsTest do
       assert Timex.diff(statistics.timestamp, time, :seconds) == 0
     end
 
-    test "returns the average time between blocks" do
+    test "returns the average time between blocks for the last 100 blocks" do
       time = DateTime.utc_now()
-      next_time = Timex.shift(time, seconds: 5)
-      insert(:block, timestamp: time)
-      insert(:block, timestamp: next_time)
+
+      insert(:block, timestamp: Timex.shift(time, seconds: -1000))
+
+      for x <- 100..0 do
+        insert(:block, timestamp: Timex.shift(time, seconds: -5 * x))
+      end
 
       assert %Statistics{
                average_time: %Duration{