Alternative high-level breakdown of patch extension algorithm

[skef]

We've already discussed changing the current description so that instead of creating a single list of patches, we create lists by invalidation status: full, partial, or no. When we discussed this the idea was to process them in that order. We've also discussed using the order of entries in the map for tie breaking, so we'll probably want to say that when the patches are broken out into these lists the ordering must be preserved.

In addition to that, I propose that we change the description to say that the list of full invalidation patches (if any) must be processed first. Then, once there are no remaining intersecting full invalidation patches, the remaining two lists can be processed concurrently. This makes the nature of partial invalidation patches clear in the algorithm and is better relative to the concerns of this thread.