Where the concern is transaction logging for crash recovery, it is natural to do the logging at the item level and to treat Table Store actions as checkpoints.
In this case, each row output action (Move, Delete, New, or Update) should create a log entry with sufficient information to reconstruct the action if necessary. Table Store actions should also be marked in the journal.
Recovery then consists of searching backward in the journal to the last Table Store transaction, opening the table as it was last stored, and then moving forward through the journal, applying the item-level changes required, and storing the updated version.
If CICS temporary storage is used for journaling, then the journal will survive a warm restart of CICS but not a cold restart.