Add ability to call fitness function in batch (e.g. once for all children in a population)

[linanqiu]

Often the fixed cost of calling the fitness function is high while the variable cost is low, especially in ML applications. e.g. calling predict on xgboost for 100 rows takes X seconds, but 100,000 rows takes around 5X seconds. I'm adding an option that allows users to write a batch fitness function and expect the entire population at once.