Shard merge recipient might read donor file data after the backup cursor is closed.

Currently DocumentSourceBackupFile cursor doesn't check if the backup cursor is alive in between batches, it only checks at the beginning of the command. So, merge recipients might read donor file data after the backup cursor is closed. For a restore using backup cursor works correctly only if the files were copied when the backup cursor is alive. In such scenarios, we expect running WT RTS  on such copied files might cause node crash/ WT instance to throw error. It's risky to always count on that WT behavior.