Parameterized loss functions

[MUST-Deep-Learning]

Some losses (e.g. weighted_cross_entropy) have parameters. These parameters might need to be set by the user (or HP sweeper). There is currently no feature to do this.

Also note that weighted_cross_entropy is handled as a bit of an anomaly in KnowIt.setup.setup_weighted_cross_entropy.

We need a framework to handle parameterized loss functions,