Change queue supervisor strategy to :one_for_all

Queue supervisors used a `:rest_for_one` strategy, which allowed the
task supervisor to keep running when a producer crashed. That allowed
duplicate long-lived jobs to run simultaneously, which is a bug in
itself, but it could also cause `attempt > max_attempts` violations.

Switching the strategy to `:one_for_all` ensures that the task
supervisor shuts down if the producer crashes.

Closes #532

lib/oban/queue/supervisor.ex

@@ -63,6 +63,6 @@ defmodule Oban.Queue.Supervisor do
       {Watchman, watch_opts}
     ]
 
-    Supervisor.init(children, strategy: :rest_for_one)
+    Supervisor.init(children, strategy: :one_for_all)
   end
 end