Update train/test splitting

[jevanilla]

We currently allow for splitting training and testing (or forecasting) data by year or fraction. To deploy the package in additional regions, some functions and their accepted configuration need an update to account for data being combined from multiple sources.

[jevanilla]

Configuration now accepts year, region and species definitions for train and test sets as of 3cc24a8

[jevanilla]

Normalization is moved to the top of transform_data() before splitting and image generation takes place in f75d275

[jevanilla]

Moved compute_gap() and check_gap() back inside make_image_list() to be called on each separate group (unique location, year, species combo) in 2f42e54