[PATCH] [Tsan] Fix the signal_recursive.cc test to pass on systems with high loads

[Viktor Kutuzov]

OK, the problem is that sometimes StartWorld() does pthread_kill(kSigRestart) before the suspend handler calls sigsuspend(). In fact, it tries to restart the world even before the first sem_post() call returns control. So what we need is some delay between StopWorld() and StartWorld(). I'm going to determine what duration would make the buildbot happy and then prepare an update.

http://reviews.llvm.org/D6504