Use contiguous index queue in bulk execution to reduce number of spawned tasks

[msimberg]

The fork-join executor uses contiguous_index_queue as a lightweight queue for work stealing between the worker threads working on items in the current bulk parallel region. We could use the queue also in execution::bulk for thread_pool_scheduler and parallel_executor. This allows work stealing to still happen but removes the need for "hierarchical scheduling" since only one task per worker thread is needed.

[msimberg]

This is now in use in thread_pool_scheduler (#5531).

[hkaiser]

I think we can close this as #5531 provides a means of using the index queue as needed.