DataTask Composition Policies and Granularity

[Keith-Bateman]

• Create a factory pattern for different composition policies
• DataTasks can be composed on a 1:1 basis (1 task for 1 operation), but in practice this doesn't require a distinct policy (default policy can subsume its functionality)
• Default policy should split operations larger than a specified maximum DataTask size into multiple DataTasks, explore whether it's better to split I/O evenly across DataTasks or to put the biggest operation first.
• Aggregating policy should still have splitting logic to maintain a maximum DT size, but also aggregate operations smaller than a given minimum size. There are complications regarding this, as different files, seeking, or changing operations can cause a naive aggregating policy to not work as intended. As a result, we will have to explore different solutions to create a more intelligent aggregating policy. A suggested baseline is to simply stop aggregation once variable conditions are detected (though certainly this isn't the best option).