[RFA:] (take 2) Make 23_containers/set/modifiers/16728.cc not time out nonnatively plus testsuite feature

[Benjamin Kosnik]

> See <URL:http://gcc.gnu.org/ml/gcc-patches/2007-02/msg00450.html>
> and discussion; this is in effect a s/!native/simulator/,
> introducing and using the new "target simulator" discriminator,
> which uses a board-specific cookie, thus more precise.

This is ok, please put it in gcc-4_2-branch as well.

Thanks for working on the timeout issues. There is an outstanding PR 
about the timeout issues: 28870. I would really like to fix this up for 
gcc-4.2.0.

As far as your questions about this testcase, here's more info:
(http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16728)

The background is: this test was added to the libstdc++ testsuite from 
the optional performance testsuite, thus the style and form of it, and 
the memory allocation pattern (ie nonlinear): it was likely brought over 
with very little modifications (although the timer was stripped out). It 
was added because this was compiler FE/BE damage without a test case, 
and broke libstdc++ performance testing for months.

So, I added this by request to make sure that something in an 
"always-run" test was breaking.

Perhaps this could be set to lower iterations everywhere. I don't have 
the broken sources checked out ATM to check.

-benjamin