Normalization layer - Swappable?

[CCranney]

Currently, the Normalization step is hard-coded for Layer Normalization, as was described in Attention Is All You Need. I've heard, however, that this too is something that can be adjusted to improve performance.

This would require more research, but I'm thinking it would be a good idea to update the SublayerUnit class so that it can swap out LayerNormalization for other normalization methods.

[CCranney]

This is under development - see the dev/expand-norm-and-activations branch

[CCranney]

This was implemented in #44.