We (may) need a specialized `train/test/validation` split method for time-series

[ezmiller]

Link with details about how this kind of thing works: https://otexts.com/fpp3/tscv.html. 👈 is just one variation. There are others.

Could also result in some scheme by which we try different variations. I.e. one way to express this is a function that returns a sequence of split-types.