Try categorical embeddings using the embed package

[dfsnow]

Currently, the CCAO uses LightGBM's native categorical handling to deal with categoricals as features. However, there may be more efficient/better ways to handle categoricals using various embeddings. See https://embed.tidymodels.org/

[dfsnow]

We tested out using embeddings of categoricals via the same process as our linear baseline model. However, the embeddings have worse performance than just letting LightGBM handle the categoricals natively. Additionally, the dependencies for the embed package are quite heavy, so as of #190 we've removed embeddings entirely (including from the linear model).