[Patch] Implement Option 3 of DR 431

[Paolo Carlini]

Paolo Carlini wrote:
> I know about that, of course, and we already discussed the issue a 
> bit. Given that otherwise container::swap is totally broken for 
> unequal allocators, and your analysis in N1599, it's the best we can 
> do... well you agree of course ;) and Gaby too. Maybe, additionally to 
> mentioning DR 431 in the docs, as we usually do by default for 
> implemented DRs, we can also add a note to the 4.2 release notes, but 
> we have got time for that. If nobody objects, I'm definitely going to 
> commit this work to mainline too.
I want to add a remark, which is for sure obvious to Howard, for 
example, but better be clear: a very, very, nice side effect of the 
optimization for empty allocators, implemented via template 
specialization, is that empty allocators, the only one we currently 
support as far as container::swap is concerned, don't see anything new 
going on, don't see allocator swap of any sort taking place.

Paolo.