[Patch] Convert pool_allocator

[Paolo Carlini]

Dhruv Matani wrote:

>I have a few things to say:
>
>1. This allocator maintains a per-type pool as opposed to a global only
>one pool for all types scheme, which I suspect was the original
>intention.
>  
>
Right. Doesn't seem a bad idea, anyway...

>2. The check in the allocate function for __n could be removed, because
>the standard does not say anything, so we may assume undefined behaviour
>if __n is 0.
>  
>
Let's keep it, was already there and makes allocate and deallocate more 
consistent.

>3. There is a line:
>if (__builtin_expect(__ret == 0, 0))
>__throw_bad_alloc();
>  
>
Same, was already there. Let's keep the old behavior for now.

Paolo.