csvstat: Add support for --no-inference

Use case:

I have strings of the form "1-1-19" and "1-1-20". I want e.g. counts of those as unique strings. They are interpreted as dates which adds massive confusion. I'd love to have both implicit inference for numbers, and no fancy guessing of "implicit dates" since the latter is so wide-ranging and convoluted in the international public mind.

From https://github.com/wireservice/csvkit/issues/935#issuecomment-396949179