Avoid QVector<T> in fast color (De-)Linearization

Constructing a QVector is expensive, so expensive that it was
actually slower than the non-fast versions.
It ranked pretty high in perf and removing it improved HSY'->RGB
conversion times quite a bit.