Formally switch to radix ordering in vec_order()

[DavisVaughan]

Closes #1302

• Renames vec_order() to new vec_order_base()
• Renames vec_order_radix() to vec_order()
• Folds order-radix.R into order.R (similar with tests, and C side)
• Big NEWS bullet outlining the changes
  • One of which is that I've added ... to both vec_order() and vec_sort(), which feels appropriate given that we have so many optional arguments to them now, and could potentially add more.

Will likely wait quite a while to merge this PR, as it will immediately affect dplyr and slider

[lionel-]

Since we are changing the default I think we should consider a C sort variant with letters swapped so that it's closer to POSIX sorting. It's probably nearly as efficient and will sort more usefully with English datasets (and other alphbetical languages).

[lionel-]

Eventually we'll have character classes that attach a locale to a vector. Compared to sorting with the current locale, these will have the advantage of producing a consistent sort across platforms. The most obvious way to implement this class is with an order proxy.

So I'm wondering what is the interaction between chr_transform, which is presented as a way of setting a locale for sorting, and the order proxy? I think it would be helpful to clarify this and maybe implement a POC localised character class before cementing the design.

[hadley]

We're going to argue that this change increases consistency with stringr and data.table; I think using some custom sort would be a step in the wrong direction.

[DavisVaughan]

Superseded by #1435 which is a clean rebase of this on master