cli: improve the csv/tsv output (and sql results in general)

[knz]

Previously the csv/tsv output would output the total row count at the beginning, now it does at the end.

@bdarnell suggests: having the row count at the end makes parsing difficult. Do we really need it?

Initially I thought that the row count was useful at least for humans so that a csv/tsv output on the screen can be easily sized (humans are not good at counting manually). So Ben suggested we only display the row count when interactive, which is indeed a simple change.

However after thinking this some more I realized that we have another need: when a client issues multiple queries one after another, we also need a delimiter to separate result batches. This is especially true for csv/tsv where otherwise there would be no delimiter and no way to distinguish the column header of a new result batch from an extra row in the previous batch.

That said, the row count is not a very good delimiter either -- it is a variable string and thus difficult to match.

So instead I propose to do the following:

• make the row count only appear on interactive shells by default (perhaps with an option to hide/show it)
• add two new options "footer" and "header", with footer set to "--" by default, to be printed before and after result batches

@bdarnell would this be adequate?

[knz]

(psql has footer/header too)

[bdarnell]

The common case is a single query; this should not print a header/footer/delimiter by default. I think a delimiter (printed in between results) makes more sense than a header/footer (printed with every result), but I'd be OK with a header/footer as long as it's only used when there are multiple queries (or maybe if the footer is a blank line - something that CSV parsers won't choke on)

[knz]

I like the blank line the best. It is visually unintrusive while being extremely reliable (you can't have a blank line as part of a field without quotes). I will do that and avoid the feature bloat altogether.