Specialize Spine for default times

The current Spine implementation tracks a list of batches, each of which can hold arbitrary updates. At the same time, we often see the bulk of updates are at times in the past which are equivalent to <T as Lattice>::min(), the smallest possible time. These "updates" are essentially describing the current dataset.

The set of updates with times equivalent to the smallest possible time have a few important properties:

1. They all have the effectively the same time. We can remove the (large) time field from the updates, economizing on storage.

2. They accumulate up to positive counts, when the ring is the integers. In this case, we can use the multiplicity of the record to indicate the diff field.

These two optimizations, the first making lots of sense and the second being useful for some data, bring the in-memory size for graph collections down to pretty close to the adjacency list representation.

To implement such a specialized spine, we need to extract the appropriate updates from the general batches, as we cannot accommodate more general updates. This means we need to write some specialized merging, and take some care when we do it.

Start by locating the Spine implementation and its general batch-merging path; the issue provides no file or test names. Trace how default-time updates can be extracted and merged separately. Done means a specialized spine handles these updates while preserving the stated storage and multiplicity optimizations.