Per-partition output ordering for ExecutionPlan

We have a leaf Scan operator that reads data from in-memory and on-disk sources. Data from in-memory sources are not ordered, while data from on-disk sources are ordered since the flush operation enforces ordering.

Data from different source would be wired to different partitions of the Scan operator. I'm wondering if it's reasonable to add partition-level output ordering in the PlanProperties for optimizing scenarios with plan like Scan -> Sort.

Start by examining the ExecutionPlan and PlanProperties concepts, then trace how Scan exposes ordering for in-memory and on-disk sources. Define whether partition-level output ordering can be represented and how the Scan -> Sort optimization would determine that the work is complete.

Written by the indexing model from the issue text.