Fix PerfTest standard deviation calculation.
https://bugs.webkit.org/show_bug.cgi?id=98115

Reviewed by Ryosuke Niwa.

Previously our standard deviation calculation was incorrect. This patch
updates perftest.py's algorithm to calculate the sample standard deviation
(with Bessel's correction) using Knuth's online algorithm:
http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Online_algorithm
An existing test has been modified to prove our new results.

This patch also updates runner.js to use Bessel's correction in
its sample standard deviation calculation, which is more accurate
for small sample sizes.

Additionally, runner.js has been modified to not calculate
the 'sum' statistic, which was not very useful.

PerformanceTests:

* resources/runner.js:
(PerfTestRunner.computeStatistics):

Tools:

* Scripts/webkitpy/performance_tests/perftest.py:

    The unused variable valueSum has also been removed.

(PageLoadingPerfTest.run):
* Scripts/webkitpy/performance_tests/perftest_unittest.py:

    This test calculates the stdev of {2000, 3000, ..., 20000} which
    was hand-calculated using a spreadsheet.

(TestPageLoadingPerfTest.test_run):

LayoutTests:

* fast/harness/perftests/perf-runner-compute-statistics-expected.txt:
* fast/harness/perftests/perf-runner-compute-statistics.html:
* fast/harness/perftests/runs-per-second-log-expected.txt:


git-svn-id: http://svn.webkit.org/repository/webkit/trunk/LayoutTests@130135 268f45cc-cd09-0410-ab3c-d52691b4dbfc