[Patch] Fix adjacent_difference in parallel-mode

[Johannes Singler]

Paolo Carlini wrote:
> Hi all, hi Johannes,
> 
> today I run again after some time check-parallel and one or more (don't
> remember now, sorry) adjacent_difference tests failed, because of typos
> introduce by the uglification, as far as I can see. 

Thanks for the fix.

> I would ask you,
> however, to go again through for_each_selectors.h, because the mix of
> first vs _M_first and second vs _M_second (see, for example
> __transform1_selector vs __transform2_selector) makes me a little
> nervous... 

I have looked into it, and consider it fine.

__transform2_selector takes an _IteratorTriple, which is a custom type,
so I have uglified _M_first, _M_second, _M_third.  __transform1_selector
takes a std::pair of iterators, so there was no need to uglify first and
second.

-- Johannes