Three small improvements

• improve thread pool utilisation in HistoFiller by shuffling the work queue so that large files go first. Avoids having a single large file at the end with unoccupied cores.
• make the SuperSmoother available in PlotUtils and adapt the corresponding populator
• remove unneeded dynamic_cast from the VirtualDerivedBranch.

Threading efficiency looks nice with this (100M JPsi events in 100 files of varying sizes, on a 12700H with 20 threads)