fix synchronization in collective DistArray initializations/transformations

[evaleev]

There are several closely-related issues around synchronization in initialization/transformation of DistArray's:

• diagonal_array uses taskq.fence() which can lead to deadlock without global synchronization (e.g. if rank receives tasks not generated by diagonal_array whose submission/execution requires progress beyond diagonal_array)
• to avoid similar problems elsewhere completion of all tasks of a given type is typically accomplished by awaiting on ntask_created == ntask_completed (e.g. https://github.com/ValeevGroup/tiledarray/blob/master/src/TiledArray/conversions/foreach.h#L286 , with ntask_created named task_count and ntask_completed named counter). As is now obvious updating ntask_completed from within the task, before its result has been assigned to the future creates a potential race, since the awaited condition will be satisfied before the effects of the task completion have been completed. The correct way to ensure local completion is to update ntask_completed using a callback attached to the future that will be assigned the task result.

To address these issues, beyond direct fixes, it would be convenient to add fencing support to collectives like DistArray::init_tiles directly. This PR accomplishes that.