Currently we publish all tables with the epoch matching the last timestamp of the storage transaction. This can lead to some tables' catalog entries with earlier timestamps being included in a checkpoint that doesn't include the tables yet. A concrete example:

• Table1 and table2 are created at ts1 and ts2
• The storage transaction commits. Both tables use ts2 for their schema epoch
• The stable timestamp advances to ts1. This can happen when log service has just acknowledged ts1, after the storage transaction has committed and tables published
• A wt checkpoint happens and includes the catalog entry of table1 written at ts1. However, since table1's schema epoch is at ts2, it's not included in the checkpoint, causing the mismatch