Investigate what may cause sphere history to become bloated

[cdata]

As of #608 , we have a test that stresses the payload size for synchronizing with the gateway. In one half of the test, we generate 1000 revisions to history that includes small changes (32 bytes each) to a random selection of ~100 slugs in a sphere. This half of the test reliably produces history that approaches or exceeds 2MB in total size when bundled.

This seems like a lot, but maybe it is to be expected given the contrived nature of the test. Regardless, we should investigate where the cost comes from and mitigate it if there are opportunities to do so.