newrt select() implementation vulnerable to scheduler/task environment race

[bblum]

As outlined in #8132.

In this case the race is around ready_index. Here is the closure:

 53     do sched.deschedule_running_task_and_then |sched, task| {
 54         let task_handles = task.make_selectable(ports.len());
 55 
 56         for (index, (port, task_handle)) in
 57                 ports.mut_iter().zip(task_handles.consume_iter()).enumerate() {
 58             // If one of the ports has data by now, it will wake the handle.
 59             if port.block_on(sched, task_handle) {
 60                 ready_index = index;
 61                 break;
 62             }
 63         }
 64     }

Since we are letting any one of the ports wake the task, the access of ready_index has the potential to clobber arbitrary values on the task's stack.

This can't be fixed just by reordering accesses, since this is in a loop, the select implementation needs to be able to write to ready_index after storing the task handle in previous ports.