This commit introduces the Top Changes page. This page retrieves
'top 10' changes for all host/platform/branch combinations from
a new 'change' database table. The script that updates this table
for a given host/platform/branch combination is run automatically
after uploading a new set of results (since this is when new
changes may potentially arise).

This commit also removes the ranking feature as this is obsoleted
(more or less) by the new feature.

The formulas for LCSS and LCSS1 are now back to the original:

LCSS1 = LCMS * LCGSS * LCLSS * LCDS1
LCSS = LCSS1 * LCDS2

A simple product turned out to be the most robust way to avoid assigning
significant scores to certain outlier cases after all.

This patch also redefines the global separation score slightly to make
it more robust (the values before the last change are now compared against
all values at or after the last change, rather than just the last change
value itself).

Finally, the default maximum durability score is adjusted from 30 to 10
(i.e. 10 seems to be sufficient to establish stability).

The formula for LCSS is adjusted from

LCSS = (LCMS + LCGSS + LCLSS + LCDS1 + LCDS2) / 5.0

to

LCSS = (LCMS + LCGSS + LCLSS + (LCDS1 * LCDS2)) / 4.0

This ensures that certain outlier cases are not falsely
given a significant score.

QCOMPARE and friends could be used to ensure that performance
was measured for the intended code path: A failed verification
step would indicate an invalid benchmark result. This information
would then be stored in the database and taken into consideration
in a later analysis.

The feature falsely relied on explicit data tags being used and
was simply ignored for test functions without such tags
(i.e. without a corresponding _data() function).

We want to track the Tieto benchmarks files as well.