hardhat 1.3.0

• New family of spruce_*_multiple() functions to support standardizing
  multi-outcome predictions (#223, with contributions from @cregouby).

• New fct_encode_one_hot() that encodes a factor as a one-hot indicator matrix
  (#215).

• default_recipe_blueprint() has gained a strings_as_factors argument, which
  is passed on to recipes::prep() (#212).

• Using a formula blueprint with indicators = "none" and character predictors
  now works properly if you provide a character column that only contains a
  single value (#213).

• Using a formula blueprint with indicators = "traditional" or
  indicators = "one_hot" and character predictors now properly enforces the
  factor levels generated by those predictors on new_data during forge()
  (#213).

• Using a formula blueprint with indicators = "none" now works correctly if
  there is a variable in the formula with a space in the name (#217).

• mold() and forge() generally have less overhead (#235, #236).

• Added more documentation about importance and frequency weights in
  ?importance_weights() and ?frequency_weights() (#214).

• New internal recompose() helper (#220).