Fix memory issue in run_jobs

[CloudNiner]

One of the changes in #804 appears to have re-created the issue where each dataset import consumes a large amount of memory (up to 8gb), preventing the regular ECS management task from being used. As a workaround, we've created a custom ECS management task that allots 8gb memory instead of the usual 2gb.

The goal of this task would be to find and fix the memory regression that would allow us to import using the regular management task again.

The original memory improvement was added in #754

[CloudNiner]

This issue bit @flibbertigibbet again when testing #854