Implement tidymodels / hardhat integration

[kbenoit]

This would follow the suggestions from quanteda/quanteda#1663 by @juliasilge:

I am fully on board with modularizing the excellent work in quanteda so that the text models are separate; I think this would make the models more accessible to a wider variety of folks involved in text analysis and would also help with dependency issues. People can only take on the dependencies that they need.

I am agnostic about whether you make this truly independent of quanteda. One option is to only Suggest quanteda, the way that tidytext does, so that the dfm methods do live in this new package but people who aren't using quanteda don't need to install it to use the package.

I would love to see tidy methods for these models live in the modeling package itself and I would be happy to collaborate on this. In general, in the tidymodels/broom ecosystem, there is some movement toward tidiers living with the models when possible, except for examples like models from utils and stats and such.

One resource you might check out as you start work on this is the hardhat package, a resource for building modeling packages. Not all of it may apply to our subject area here, but I believe some will and it can help build more composable, extensible models.