zero-alloc cache (was Re: [v3] Fix PR libstdc++/10276)

[Jerry Quinn]

B. Kosnik writes:
 > I don't know. 
 > 
 > Jerry, I'm back from wandering around in the desert of the
 > testsuite work, so I can help on this too. I'm not quite sure what's up.
 > 
 > My plan now is to read/figure out what is up with the exception issue as
 > discussed previously in the meantime.
 > 
 > -benjamin

I've been stuck in the real world for a while, but I'm now looking at
this again.  Then I'll get back to reworking the caches to be indexed
by charT for a locale.

I just tried on an Athlon with the current 3.3 branch and valgrind
1.9.5 is not showing any leaks.  Very odd ...  The caches for the
stdio objects should still leak, no?  I have to go back and check on
the other machine.

Jerry