forderv should be able to deal with more than one data.table efficiently.

[arunsrinivasan]

Currently forder takes a single atomic vector, list or data.table as input. But it'd be much useful for so many other internal functions if it'd also accept a list of data.tables.

If that were possible, for example, unique.data.table's function definition could be changed to:

unique.data.table <- function(x, ..., incomparables =,  ) {
    x = list(x, ...) # in R v3.1 or using data.table internal functions to not copy the list
    # rest of the code
}

This allows us to optionally obtaining the unique of a list of data.tables without having to rbind them first.

Even more, this is extremely useful for all the setops functions, issue #547. Specifically, this line - where crux of the computations happen could be made much more (memory + speed) efficient!

Also could be useful in implementing #873.