As no one replied to this, I assumed it probably wasn’t a big deal - so I went ahead and tried to fix the out-of-sync table using pt-table-sync with the --nocheck-slave option, and master and slave defined using DSNs. And I’m delighted to report that it seems to have worked fine. (One piece of luck was that the table is very rarely updated, so there was little chance of problems when the slave threads were restarted. I’m not sure that it would have gone so well had the table been subject to heavy updating.)
(If anyone who sees this is involved with maintaining the documentation, I’d suggest that the bit of the pt-table-sync manual mentioned above is changed to say that the refusal to work with a master/slave pair can be overridden with the --nocheck-slave option. With appropriate caveats, of course …)