Consider comparing with `tidyverts`

The `tsibble` package looks polished for working with time-indexed data. We have geo and time-indexed data, which complicates things a bit, but would be nice to reuse their work.

Discussion already happened in #7, this issue might be retreading old ground.

But other than that, there is also the `fable` package, which might be useful for some ideas in epipredict.