Ignore sysmsg tgkill errors if the sysmsg thread is already dead.

This can happen in case the entire sandbox is being torn down
unceremoniously such that the order of who dies first isn't controlled
for.

I am not sure if this should also set `sc.subprocess.dead`, as the
`tgkill` seems to be specifically about a sysmsg thread associated
with this subprocess, not about the subprocess itself.

PiperOrigin-RevId: 750384169

Author: EtiennePerot

pkg/sentry/platform/systrap/shared_context.go

@@ -141,7 +141,7 @@ func (sc *sharedContext) NotifyInterrupt() {
 	}
 
 	t := sysmsgThread.thread
-	if e := hostsyscall.RawSyscallErrno(unix.SYS_TGKILL, uintptr(t.tgid), uintptr(t.tid), uintptr(platform.SignalInterrupt)); e != 0 {
+	if e := hostsyscall.RawSyscallErrno(unix.SYS_TGKILL, uintptr(t.tgid), uintptr(t.tid), uintptr(platform.SignalInterrupt)); e != 0 && e != unix.ESRCH {
 		panic(fmt.Sprintf("failed to interrupt the child process %d: %v", t.tid, e))
 	}
 }