This simply persists tasks with priority in another task type. And let the scheduler
runner and listener deals with them as other task types does (avoid some otherwise not
so fancy computations [2]).
This implies that the dedicated task type exists in the scheduler backend [1].
Instead of doing some not so fancy computations in the scheduler runner. Simply
detect and persist task into a dedicated high task type.
That way, future save code now requests will be in their own queue and consumed by
dedicated workers [3]. Existing save code now requests will eventually be consumed as it is
today (if any remained not done yet).
Related to T3084
[1] D5488
[2] D5488#139373 (described the detail of what that would entail otherwise)
[3] D5486