PR cleanup re locale caches

[B. Kosnik]

>9828 segfault in num_put<wchar_t>::put

close, check in test cases

>10276 ostringstream mem leak

close

>Finally, 7076 is now much improved, but not 100% solved.  Mainline is
>currently about 10% slower than 2.95.  How much of a catch-all should
>this bug be?  For instance, should it be an umbrella for getting
>caching in place for facets beside numpunct?

This is now in the performance testsuite, so you could close it if you
wanted to (and the person submitting it agrees). However, I'm with
Nathan on this one: we might as well keep it open and keep on working on
optimizations.

-benjamin