Release: 0.5.1

• Added a conserve_memory setting (from chrisjen's new option for
  comparing workers), for an experiment's section or for every experiment in
  the "general" section. An experiment that conserves memory scores each
  combination as soon as it is tried and keeps only the best whole dataset
  so far, rather than every combination's dataset until the last one is
  done, which ran out of memory with many combinations of a large dataset.
  Its comparison table, scores, and branches are the same, but only the
  winner's branch has its whole dataset, so a regression_table has only
  the winner's model. Experiment.outcomes has the outcome of each
  combination (its label and group columns, predictions, probabilities, and
  metrics). Requires chrisjen 0.4.0 or later.