`order` could utilize existing index

[jangorecki]

d = data.table(x=2:1)
d[order(x, na.last=FALSE)]

[jangorecki]

This issue could be automatically handled by #4386 but the problem is that forder strips out all attributes before even calling forderv. This happens because when we call

d = data.table(x=2:1)
d[order(x, na.last=FALSE)]

We technically call forder on integer vector x, not on a d data.table. To not duplicate the work, to address that, it is best to rely on #4386 and we have to make better input massaging when preparing arguments for forderv, retaining key/index attributes.
According to ideas in #4346 the proper way to address that would be to optimize from [.data.table directly to forderv, and not to forder.

[jangorecki]

It should be possible to optimize as well d[order(x)] in cases where x does not have NAs, which can be a cheap to check if index exists.