benchmark: calibrate length of util.diff

500 + 1000 already cover the curve. 2000 doesn’t add
new qualitative insight — it just extends the same curve
further down (another ~3–4× slowdown).

According to nodejs/performance#186
this benchmark takes one minute to conclude a single run.
This should fix it.

benchmark/util/diff.js

@@ -5,7 +5,7 @@ const common = require('../common');
 
 const bench = common.createBenchmark(main, {
   n: [1e3],
-  length: [1e3, 2e3],
+  length: [500, 1000],
   scenario: ['identical', 'small-diff', 'medium-diff', 'large-diff'],
 });