Testing parallel mode and PR34106

[Paolo Carlini]

Hi,

some additional details, almost obvious, but anyway.
> FAIL: 25_algorithms/partial_sort/moveable.cc (test for excess errors)
>   
This one is because the parallel version of partial_sort doesn't move.
> FAIL: 25_algorithms/sort/vectorbool.cc (test for excess errors)
>   
No idea, we should probably look into it as soon as possible.
> FAIL: ext/mt_allocator/deallocate_local_thread-5.cc execution test
> FAIL: ext/mt_allocator/deallocate_local_thread-7.cc execution test
>   
This one happens on other arches (eg powerpc-darwin) also in normal
mode. In fact, ultimately because of this kind of failure we haven't
been able to change unconditionally mt_alloc to explicitely free all the
allocated memory before program end. That is, for some reason, checking
for the availability of atexit isn't enough to know whether the target
can really support the explicit free of the allocate memory. Or the
atexit check isn't tight enough. We should check that.

Paolo.