`tidymodels` and panel data, postprocessing, multiple outcomes

This is a long-term question / tracking.

As we continue to iterate on `epipredict`, it is becoming increasingly clear that we are implementing this framework not just for epidemiological data, but for general panel data along with other bells and whistles. One can imagine simply wanting to use `tidymodels` with panel data, and `epipredict`/`epiprocess` provides a framework to do so: convert your data to an `epi_df`, then use an `epi_workflow` rather than a `workflows::workflow`.

1. At minimum, we should demonstrate this with a vignette.
2. Down stream, we should consider whether to strip out some bits of `epipredict` and try to incorporate directly into `tidymodels`.

* The panel data request has been [around for a while](https://github.com/tidymodels/recipes/issues/166) but is not a priority.
* Neither is postprocessing predictions: https://github.com/tidymodels/workflows/issues/154 predictions.
* Neither is prediction of multiple outcomes simultaneously: https://github.com/tidymodels/workflows/issues/4