Ironically, the Stream algorithm I wrote as the simplest possible syncing algorithm does what the much more efficient algorithm I wrote some time ago can't do: sync a table without a primary key, as long as there are no duplicate rows. In fact, it's so dumb, it will happily sync any table, even if there are no indexes.
The flash of inspiration I had on Friday has turned out to be good...
My initial plans got waylaid! I didn't pull out the checksumming code first, because the code wasn't at all as I remembered it. Instead, I began writing code to handle the more abstract problem of accepting two sets of rows, finding the differences, and doing something with them. I'm ending up with a little more complicated system than I thought I would. However, it's also significantly simpler in some ways. Instead of just passing references to subroutines to use as callbacks, I'm object-ifying the entire synchronization concept...