parallel_for optimization in memory allocation

[aconstlink]

The parallel_for is using schedule( motor::shared( task_t...) ) ) in parallel_for.hpp which triggers a memory allocation with the task. When the task is done, it even triggers a deallocation. I think that is completely unnecessary!

There should be a better way to allocate tasks for the parallel_for so that tasks are directly reused. There is the memory arena which does exactly that, but then we can not route the parallel_for tasks the ordinary task scheduling way.

[aconstlink]

It came to my mind, that it could also be possible to just up the ref count of all used tasks using motor::share once more, and store all parallel_for tasks internally until destruction and just reuse then that way.

✨ That would be absolutely transparent with the general purpose memory manager currently in use! 🎉