Finish testing the transfer benchmark

[deontologician]

• .test_set - high priority
  • returns the correct fold
  • returns no entries from the training set
• .training_set - high priority
  • returns all entries except from the test set
• .default_dir - low priority
  • check that it returns something matching a regex
• .game_agent_filename - low priority
  • check with a regex
• .fold_agent_filename - low priority
  • check with a regex
• .tested_agent_filename - low priority
  • check with a regex
• .ensure_game_agents - medium priority
  • if there's no agent for a game it makes one
  • if there is an agent, it does nothing
• .test - low priority
  • creates a TestRun
• .train - low priority
  • creates a TrainingRun
• .do_folds - high priority
  • all fold_agents get trained
  • fold agents are saved after training
  • a fold agent gets tested on each test game
  • a fold agent results on a game are saved
  • the "tested_agent' is saved