decide/document stance on partially trained workflows

[simonpcouch]

re:

workflows/R/post-action-tailor.R

Lines 128 to 135 in 9e6ca98

	# note that the preprocessor and model fit don't need to be "untrained"
	# with new_stage_* since they are unaffected by the post-processor.
	new_workflow(
	pre = x$pre,
	fit = x$fit,
	post = new_stage_post(actions = actions),
	trained = FALSE
	)

In the PR thread:

ec0effa surfaces an important point; removing/updating a postprocessor from an otherwise trained workflow need not remove the preprocessor and model fits, as they won't be affected by the removal of the postprocessor. This introduces the possibility of a "partially trained" workflow, where a workflow with trained preprocessor and model but untrained postprocessor should be able to fit without issue.

[simonpcouch]

This was already the case with preprocessors and models before. We'll continue only supporting partially-trained workflows via the @keywords internal .fit_*() functions and document as such.