sys_tgkill: hand the syscall to the kernel in the standard way, rather

sys_tgkill: hand the syscall to the kernel in the standard way, rather
than doing it inline.  Doing it inline screws up on ppc32-linux if
we're sending an async signal to ourselves (the same thread) because
the kernel immediately hands the signal to async_sighandler() which
then dies at the assertion that this thread's state is VgTs_WaitSys.
From which I conclude this wrapper has always had a race against the
kernel which did not show up on x86 or amd64.  (and/or that I don't
understand this stuff too well)

git-svn-id: svn://svn.valgrind.org/valgrind/trunk@5169