Merge speed test side dud track into raimo/stdlib/rand-polish

    * mwc59 and dedicated shuffle implementations:
      Revert "Try the speed of mwc59 in particular for shuffle"
      Revert "Try dedicated shuffle_mwc59"
      Revert "Try dedicated shuffle_exsss"
      Try dedicated shuffle_exsss
      Try dedicated shuffle_mwc59
      Try the speed of mwc59 in particular for shuffle

    Results:
    * mwc59 only makes a difference for shuffle, not for bytes
    * An inlined shuffle_mwc59 is about 20 % faster than exsss
    * An inlined shuffle_exsss is as fast as shuffle_mwc59

    Hence, it can give 20% to inline the PRNG functions in shuffle,
    so it is not worth the code duplication, and the speed of mwc59
    is buried in the shuffle algorithm so there is no point in
    having it in the plug-in framework.

Author: RaimoNiskanen