changefeedccl: resume after truncate looks like it repeatedly hits the same boundary

On master:

```sql
CREATE TABLE foo...
INSERT INTO foo ...
CREATE CHANGEFEED FOR foo INTO 'null://sink' WITH on_error='pause';
-- wait for changefeed to checkpoint
TRUNCATE TABLE foo;
-- Changefeed pauses, correctly.
RESUME JOB ;
-- Changefeed claims to be running
INSERT INTO foo...
-- Changefeed pauses again with the same error
```

Motivated by [this support issue](https://github.com/cockroachlabs/support/issues/2127), but it seems likely we have multiple issues around resuming a job on a truncated table (or any table whose primary index has changed?).

Jira issue: CRDB-25302

Epic CRDB-28883