Remove `predict_<type>_glmnet()` for all types but `"raw"`

[hfrick]

Following on from #877, the glmnet methods for predict_numeric(), predict_class(), and predict_classprob() would only evaluate the model spec before handing off to the method for model_fit. Let's check if that is necessary or if we can remove those functions altogether.

[hfrick]

parsnip/R/glmnet-engines.R

Lines 48 to 61 in 3cb0914

	predict_numeric_glmnet <- function(object, new_data, ...) {
	object$spec <- eval_args(object$spec)
	predict_numeric.model_fit(object, new_data = new_data, ...)
	}
	predict_class_glmnet <- function(object, new_data, ...) {
	object$spec <- eval_args(object$spec)
	predict_class.model_fit(object, new_data = new_data, ...)
	}
	predict_classprob_glmnet <- function(object, new_data, ...) {
	object$spec <- eval_args(object$spec)
	predict_classprob.model_fit(object, new_data = new_data, ...)
	}

[hfrick]

Looks like we may not need eval_args() anymore. Current plan:

• demonstrate that we indeed do not need eval_args() anymore
• remove usage of this in extension pkgs and parsnip
• deprecate eval_args()
• remove predict_<type>_glmnet() where possible