Yes, the work-around in the meantime is to disable the S&F disk caches, rename/delete the corresponding folders in {installdir}/data/datacache, then re-enable the disk caches. I skipped the last step, as the client's system doesn't need disk caching. (Generously provisioned, lightly loaded.) The trend gaps are gone.
The remaining issue, the attempt to insert null into the DB during a MemoryForwardTransaction, is still happening but does not appear to be causing any ill effect. There's nothing to look at with kindling, as there are no quarantined records now. The log entry suggests that the error is in a metadata table, not a history partition.
{ I closed the support ticket, as the client-visible problem is fixed. }