Allow 2d array input in CountVectorizer?

I wonder if we should allow (n_samples, 1) input to CountVectorizer. I think we had this discussion before, right now we basically have a special case in ColumnTransformer with scalar columns to produce 1d arrays.
Does anyone remember where this discussion was?

A lot of my students are struggling with putting a CountVectorizer into a ColumnTransformer and it might be nice to be more generous with the input formats, and ensure we raise nice error messages.

import pandas as pd
from sklearn.feature_extraction.text import CountVectorizer
X = pd.DataFrame(["what is my purpose?", "You bring the butter."])
CountVectorizer().fit_transform(X)

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-6-6cae33e8c706> in <module>
      3 from sklearn.compose import ColumnTransformer
      4 X = pd.DataFrame(["what is my purpose?", "You bring the butter."])
----> 5 CountVectorizer().fit_transform(X)

~/miniconda3/lib/python3.7/site-packages/sklearn/feature_extraction/text.py in fit_transform(self, raw_documents, y)
   1218 
   1219         vocabulary, X = self._count_vocab(raw_documents,
-> 1220                                           self.fixed_vocabulary_)
   1221 
   1222         if self.binary:

~/miniconda3/lib/python3.7/site-packages/sklearn/feature_extraction/text.py in _count_vocab(self, raw_documents, fixed_vocab)
   1129         for doc in raw_documents:
   1130             feature_counter = {}
-> 1131             for feature in analyze(doc):
   1132                 try:
   1133                     feature_idx = vocabulary[feature]

~/miniconda3/lib/python3.7/site-packages/sklearn/feature_extraction/text.py in _analyze(doc, analyzer, tokenizer, ngrams, preprocessor, decoder, stop_words)
    101     else:
    102         if preprocessor is not None:
--> 103             doc = preprocessor(doc)
    104         if tokenizer is not None:
    105             doc = tokenizer(doc)

~/miniconda3/lib/python3.7/site-packages/sklearn/feature_extraction/text.py in _preprocess(doc, accent_function, lower)
     66     """
     67     if lower:
---> 68         doc = doc.lower()
     69     if accent_function is not None:
     70         doc = accent_function(doc)

AttributeError: 'int' object has no attribute 'lower'

Start with CountVectorizer.fit_transform and _count_vocab in the feature-extraction text implementation, then compare the scalar-column handling in ColumnTransformer. Clarify which (n_samples, 1) inputs are supported and what errors should be reported, and add regression coverage for the accepted and rejected forms.