Switch to hardhat's version of contr_one_hot()

[hfrick]

closes #1215, as a pair with tidymodels/hardhat#278

[hfrick]

That failed check is one on the parsnip overhead but since that test passes on all other platforms, I think it's not related to this change.

[topepo]

Looks good. I ran revdeps and there was a good test case in the LorenzRegression package (using parsnip::contr_one_hot())

[topepo]

Although I do see an additional warning for

Found the following significant warnings:
Warning: replacing previous import ‘hardhat::contr_one_hot’ by
‘parsnip::contr_one_hot’ when loading ‘viralmodels’

I'll add an issues to their repo.