Uses a reference to the collator instead of copying and reinitializing it again and again. This is the reason for the speedup.
Changing the implementation from a Functor class to a Lambda removes some boilerplate code, but is not relevant for performance.
This requires a workaround for https://bugreports.qt.io/browse/QTBUG-69361
Just a single comparison to force the clean state of QCollator.