[RFC] Removal of allocator::construct / destroy vs our containers

[Benjamin Kosnik]

>> PS: IMO, in case we end keeping those allocator members and, as a 
>> consequence, we keep on calling them from the containers, we have to 
>> improve the dispatchings in stl_construct.h and stl_uninitialized.h to 
>> avoid allocators by value. Seems doable, after all, only more complex...

Agreed, this would be nice.

> For concreteness, the below appear to work, regetests fine 
> (unfortunately some tweaks to rope are needed, but otherwise the 
> infrastructure in place seems already ok...)

Cool! I guess this waits for a more elaborate discussion in Toronto as 
well...

-benjamin