Maybe make task-local storage work from run_in_worker_thread functions

[njsmith]

This is handy for logging, and safe because we know that either the task or the thread is running, but never both - so no locking is needed. But maybe confusing (it can't work for run-locals, or run_in_worker_process)?

What should current_task() return in a run_in_worker_thread function? Currently we effectively find the task-locals by looking up the current task, so the most naive way to make task-locals work would to make current_task() work and then task-locals would work for free. Is that good or bad?