Dropping numpy array support for TableVectorizer

[Vincent-Maladiere]

Should we stop supporting numpy array inputs for TableVectorizer?

• From a data scientist's UX point of view, numpy arrays are seldom used as inputs to stats or ML models. Data scientists mainly use dataframes, from fetching data to writing data.
• Conceptually speaking, TableVectorizer deals with columns of different dtypes, while a numpy array only has one.
• This would allow us to avoid dealing with column indices instead of column names.
• This would also enable long-term laziness and support for different dataframes.

WDYT?

[GaelVaroquaux]

Let's do it for TableVectorizer (but with a clean error for inputs that are not understood)

[jeromedockes]

this has been fixed by #902 -- if the input is a numpy array a warning is issued and it is converted to a dataframe