When `epi_recipe()` outputs a regular `recipe`, errors can be confusing

If `epi_recipe()` is provided data that's not an `epi_df`, it will output a regular `recipe` rather than an `epi_recipe`, causing some step functions to output somewhat confusing messages.

Artificial example:

``` r
library(epipredict)
#> Loading required package: epiprocess
#>
#> Attaching package: 'epiprocess'
#> The following object is masked from 'package:stats':
#>
#> filter
#> Loading required package: parsnip
library(tibble)
epi_recipe(as_tibble(case_death_rate_subset)) %>%
step_epi_ahead(death_rate, ahead = 7) %>%
step_epi_lag(death_rate, lag = c(0,7,14))
#> Error in `step_epi_ahead()`:
#> ! This recipe step can only operate on an `epi_recipe`.
```

^{Created on 2023-04-26 by the [reprex package](https://reprex.tidyverse.org) (v2.0.1)}

More realistically, @dshemetov and I ran into this when feeding in a `covidcast_signal` before it had been converted to `epi_df` format.

A couple quick ideas on potential resolutions:
- Add S3 methods to the steps for regular recipes, just to give better error messages.
- Always output an `epi_recipe` from `epi_recipe()`, but complain in step functions if the appropriate roles have not been set. This might require fixing some hard-coded references to `geo_value` and/or `time_value` columns (rather than selecting based on roles) in some steps.