API/DOC: Specification for `names` parameter in read_csv

[gfyoung]

On master (04356a), the documentation is unclear as to what a valid names parameter is for read_csv.

• Does it need to have the same length as the number of columns in the file? OR
• Does it need to have the same length as usecols if specified?

I find evidence in the test cases to support both (C engine allows both for example, but the Python engine disallows the former). This point should also be clarified in the documentation.

xref #14671.

[gfyoung]

I suspect that I would lean towards using the latter, since why specify all of the column names if you're not going to use all of them?

[jreback]

is this covered?

[gfyoung]

@jreback : What do you mean by this? This is just a behavior question that I'm still unsure about but was hoping to get some feedback before making any decisions.

[gfyoung]

Marking this as a bug because:

• It works on the C engine (but not the Python engine)
• This ambiguity doesn't exist when there is no existing header. It works both ways for both engines.