add hard_cross_entropy

[coreylowman]

Current only works for actual probability distributions. hard cross entropy only has 1 non zero entry in inner dimension, so sum across that before taking mean

[coreylowman]

After #12, hard could accept a tensor with dtype = usize. For now, cross_entropy could accept enum { HardTargets, SoftTargets } to differentiate?

[coreylowman]

This is not really needed - the same function can be used for both, the user should just use one hot encode the class labels into probabilities before calling